The €390 million fine that the Irish Data Protection Commission (DPC) levied last week against Meta marks both the latest skirmish in the ongoing regulatory war on the use of data by private firms, as well as a major blow to the ad-driven business model that underlies most online services.
More specifically, the DPC was forced by the European Data Protection Board (EDPB) to find that Meta violated the General Data Protection Regulation (GDPR) when it relied on its contractual relationship with Facebook and Instagram users as the basis to employ user data in personalized advertising.
Meta still has other bases on which it can argue it relies in order to make use of user data, but a larger issue is at-play: the decision’s findings both that making use of user data for personalized advertising is not “necessary” between a service and its users and that privacy regulators are in a position to make such an assessment.
More broadly, the case also underscores that there is no consensus within the European Union on the broad interpretation of the GDPR preferred by some national regulators and the EDPB.
The DPC Decision
The core disagreement between the DPC and Meta, on the one hand, and some other EU privacy regulators, on the other, is whether it is lawful for Meta to treat the use of user data for personalized advertising as “necessary for the performance of” the contract between Meta and its users. The Irish DPC accepted Meta’s arguments that the nature of Facebook and Instagram is such that it is necessary to process personal data this way. The EDPB took the opposite approach and used its powers under the GDPR to direct the DPC to issue a decision contrary to DPC’s own determination. Notably, the DPC announced that it is considering challenging the EDPB’s involvement before the EU Court of Justice as an unlawful overreach of the board’s powers.
In the EDPB’s view, it is possible for Meta to offer Facebook and Instagram without personalized advertising. And to the extent that this is possible, Meta cannot rely on the “necessity for the performance of a contract” basis for data processing under Article 6 of the GDPR. Instead, Meta in most cases should rely on the “consent” basis, involving an explicit “yes/no” choice. In other words, Facebook and Instagram users should be explicitly asked if they consent to their data being used for personalized advertising. If they decline, then under this rationale, they would be free to continue using the service without personalized advertising (but with, e.g., contextual advertising).
Notably, the decision does not mandate a particular contractual basis for processing, but only invalidates “contractual necessity” for personalized advertising. Indeed, Meta believes it has other avenues for continuing to process user data for personalized advertising while not depending on a “consent” basis. Of course, only time will tell if this reasoning is accepted. Nonetheless, the EDBP’s underlying animus toward the “necessity” of personalized advertising remains concerning.
What Is ‘Necessary’ for a Service?
The EDPB’s position is of a piece with a growing campaign against firms’ use of data more generally. But as in similar complaints against data use, the demonstrated harms here are overstated, while the possibility that benefits might flow from the use of data is assumed to be zero.
How does the EDPB know that it is not necessary for Meta to rely on personalized advertising? And what does “necessity” mean in this context? According to the EDPB’s own guidelines, a business “should be able to demonstrate how the main subject-matter of the specific contract with the data subject cannot, as a matter of fact, be performed if the specific processing of the personal data in question does not occur.” Therefore, if it is possible to distinguish various “elements of a service that can in fact reasonably be performed independently of one another,” then even if some processing of personal data is necessary for some elements, this cannot be used to bundle those with other elements and create a “take it or leave it” situation for users. The EDPB stressed that:
This assessment may reveal that certain processing activities are not necessary for the individual services requested by the data subject, but rather necessary for the controller’s wider business model.
This stilted view of what counts as a “service” completely fails to acknowledge that “necessary” must mean more than merely technologically possible. Any service offering faces both technical limitations as well as economic limitations. What is technically possible to offer can also be so uneconomic in some forms as to be practically impossible. Surely, there are alternatives to personalized advertising as a means to monetize social media, but determining what those are requires a great deal of careful analysis and experimentation. Moreover, the EDPB’s suggested “contextual advertising” alternative is not obviously superior to the status quo, nor has it been demonstrated to be economically viable at scale.
Thus, even though it does not strictly follow from the guidelines, the decision in the Meta case suggests that, in practice, the EDPB pays little attention to the economic reality of a contractual relationship between service providers and their users, instead trying to carve out an artificial, formalistic approach. It is doubtful whether the EDPB engaged in the kind of robust economic analysis of Facebook and Instagram that would allow it to reach a conclusion as to whether those services are economically viable without the use of personalized advertising.
However, there is a key institutional point to be made here. Privacy regulators are likely to be eminently unprepared to conduct this kind of analysis, which arguably should lead to significant deference to the observed choices of businesses and their customers.
Conclusion
A service’s use of its users’ personal data—whether for personalized advertising or other purposes—can be a problem, but it can also generate benefits. There is no shortcut to determine, in any given situation, whether the costs of a particular business model outweigh its benefits. Critically, the balance of costs and benefits from a business model’s technological and economic components is what truly determines whether any specific component is “necessary.” In the Meta decision, the EDPB got it wrong by refusing to incorporate the full economic and technological components of the company’s business model.
Under a draft “adequacy” decision unveiled today by the European Commission, data-privacy and security commitments made by the United States in an October executive order signed by President Joe Biden were found to comport with the EU’s General Data Protection Regulation (GDPR). If adopted, the decision would provide a legal basis for flows of personal data between the EU and the United States.
This is a welcome development, as some national data-protection authorities in the EU have begun to issue serious threats to stop U.S.-owned data-related service providers from offering services to Europeans. Pending more detailed analysis, I offer some preliminary thoughts here.
Decision Responds to the New U.S. Data-Privacy Framework
The Commission’s decision follows the changes to U.S. policy introduced by Biden’s Oct. 7 executive order. In its July 2020 Schrems II judgment, the EU Court of Justice (CJEU) invalidated the prior adequacy decision on grounds that EU citizens lacked sufficient redress under U.S. law and that U.S. law was not equivalent to “the minimum safeguards” of personal data protection under EU law. The new executive order introduced redress mechanisms that include creating a civil-liberties-protection officer in the Office of the Director of National Intelligence (DNI), as well as a new Data Protection Review Court (DPRC). The DPRC is proposed as an independent review body that will make decisions that are binding on U.S. intelligence agencies.
The old framework had sparked concerns about the independence of the DNI’s ombudsperson, and what was seen as insufficient safeguards against external pressures that individual could face, including the threat of removal. Under the new framework, the independence and binding powers of the DPRC are grounded in regulations issued by the U.S. Attorney General.
To address concerns about the necessity and proportionality of U.S. signals-intelligence activities, the executive order also defines the “legitimate objectives” in pursuit of which such activities can be conducted. These activities would, according to the order, be conducted with the goal of “achieving a proper balance between the importance of the validated intelligence priority being advanced and the impact on the privacy and civil liberties of all persons, regardless of their nationality or wherever they might reside.”
Will the Draft Decision Satisfy the CJEU?
With this draft decision, the European Commission announced it has favorably assessed the executive order’s changes to the U.S. data-protection framework, which apply to foreigners from friendly jurisdictions (presumed to include the EU). If the Commission formally adopts an adequacy decision, however, the decision is certain to be challenged before the CJEU by privacy advocates. In my preliminary analysis after Biden signed the executive order, I summarized some of the concerns raised regarding two aspects relevant to the finding of adequacy: proportionality of data collection and availability of effective redress.
Opponents of granting an adequacy decision tend to rely on an assumption that a finding of adequacy requires virtually identical substantive and procedural privacy safeguards as required within the EU. As noted by the European Commission in the draft decision, this position is not well-supported by CJEU case law, which clearly recognizes that only “adequate level” and “essential equivalence” of protection are required from third-party countries under the GDPR.
To date, the CJEU has not had to specify in greater detail precisely what, in their view, these provisions mean. Instead, the Court has been able simply to point to certain features of U.S. law and practice that were significantly below the GDPR standard (e.g., that the official responsible for providing individual redress was not guaranteed to be independent from political pressure). Future legal challenges to a new Commission adequacy decision will most likely require the CJEU to provide more guidance on what “adequate” and “essentially equivalent” mean.
In the draft decision, the Commission carefully considered the features of U.S. law and practice that the Court previously found inadequate under the GDPR. Nearly half of the explanatory part of the decision is devoted to “access and use of personal data transferred from the [EU] by public authorities in the” United States, with the analysis grounded in CJEU’s Schrems II decision. The Commission concludes that, collectively, all U.S. redress mechanisms available to EU persons:
…allow individuals to have access to their personal data, to have the lawfulness of government access to their data reviewed and, if a violation is found, to have such violation remedied, including through the rectification or erasure of their personal data.
The Commission accepts that individuals have access to their personal data processed by U.S. public authorities, but clarifies that this access may be legitimately limited—e.g., by national-security considerations. Unlike some of the critics of the new executive order, the Commission does not take the simplistic view that access to personal data must be guaranteed by the same procedure that provides binding redress, including the Data Protection Review Court. Instead, the Commission accepts that other avenues, like requests under the Freedom of Information Act, may perform that function.
Overall, the Commission presents a sophisticated, yet uncynical, picture of U.S. law and practice. The lack of cynicism, e.g., about the independence of the DPRC adjudicative process, will undoubtedly be seen by some as naïve and unrealistic, even if the “realism” in this case is based on speculations of what might happen (e.g., secret changes to U.S. policy), rather than evidence. Given the changes adopted by the U.S. government, the key question for the CJEU will be whether to follow the Commission’s approach or that of the activists.
What Happens Next?
The draft adequacy decision will now be scrutinized by EU and national officials. It remains to be seen what will be the collective recommendation of the European Data Protection Board and of the representatives of EU national governments, but there are signs that some domestic data-protection authorities recognize that a finding of adequacy may be appropriate (see, e.g., the opinion from the Hamburg authority).
It is also likely that a significant portion of the European Parliament will be highly critical of the decision, even to the extent of recommending not to adopt it. Importantly, however, none of the consulted bodies have formal power to bind the European Commission on this question. The whole process is expected to take at least several months.
European Union officials insist that the executive order President Joe Biden signed Oct. 7 to implement a new U.S.-EU data-privacy framework must address European concerns about U.S. agencies’ surveillance practices. Awaited since March, when U.S. and EU officials reached an agreement in principle on a new framework, the order is intended to replace an earlier data-privacy framework that was invalidated in 2020 by the Court of Justice of the European Union (CJEU) in its Schrems IIjudgment.
This post is the first in what will be a series of entries examining whether the new framework satisfies the requirements of EU law or, as some criticsargue, whether it does not. The critics include Max Schrems’ organization NOYB (for “none of your business”), which has announced that it “will likely bring another challenge before the CJEU” if the European Commission officially decides that the new U.S. framework is “adequate.” In this introduction, I will highlight the areas of contention based on NOYB’s “first reaction.”
The overarching legal question that the European Commission (and likely also the CJEU) will need to answer, as spelled out in the Schrems II judgment, is whether the United States “ensures an adequate level of protection for personal data essentially equivalent to that guaranteed in the European Union by the GDPR, read in the light of Articles 7 and 8 of the [EU Charter of Fundamental Rights]” Importantly, as Theodore Christakis, Kenneth Propp, and Peter Swire point out, “adequate level” and “essential equivalence” of protection do not necessarily mean identical protection, either substantively or procedurally. The precise degree of flexibility remains an open question, however, and one that the EU Court may need to clarify to a much greater extent.
Proportionality and Bulk Data Collection
Under Article 52(1) of the EU Charter of Fundamental Rights, restrictions of the right to privacy must meet several conditions. They must be “provided for by law” and “respect the essence” of the right. Moreover, “subject to the principle of proportionality, limitations may be made only if they are necessary” and meet one of the objectives recognized by EU law or “the need to protect the rights and freedoms of others.”
As NOYB has acknowledged, the new executive order supplemented the phrasing “as tailored as possible” present in 2014’s Presidential Policy Directive on Signals Intelligence Activities (PPD-28) with language explicitly drawn from EU law: mentions of the “necessity” and “proportionality” of signals-intelligence activities related to “validated intelligence priorities.” But NOYB counters:
However, despite changing these words, there is no indication that US mass surveillance will change in practice. So-called “bulk surveillance” will continue under the new Executive Order (see Section 2 (c)(ii)) and any data sent to US providers will still end up in programs like PRISM or Upstream, despite of the CJEU declaring US surveillance laws and practices as not “proportionate” (under the European understanding of the word) twice.
It is true that the Schrems II Court held that U.S. law and practices do not “[correlate] to the minimum safeguards resulting, under EU law, from the principle of proportionality.” But it is crucial to note the specific reasons the Court gave for that conclusion. Contrary to what NOYB suggests, the Court did not simply state that bulk collection of data is inherently disproportionate. Instead, the reasons it gave were that “PPD-28 does not grant data subjects actionable rights before the courts against the US authorities” and that, under Executive Order 12333, “access to data in transit to the United States [is possible] without that access being subject to any judicial review.”
CJEU case law does not support the idea that bulk collection of data is inherently disproportionate under EU law; bulk collection may be proportionate, taking into account the procedural safeguards and the magnitude of interests protected in a given case. (For another discussion of safeguards, see the CJEU’s decision in La Quadrature du Net.) Further complicating the legal analysis here is that, as mentioned, it is far from obvious that EU law requires foreign countries offer the same procedural or substantive safeguards that are applicable within the EU.
Effective Redress
The Court’s Schrems II conclusion therefore primarily concerns the effective redress available to EU citizens against potential restrictions of their right to privacy from U.S. intelligence activities. The new two-step system proposed by the Biden executive order includes creation of a Data Protection Review Court (DPRC), which would be an independent review body with power to make binding decisions on U.S. intelligence agencies. In a comment pre-dating the executive order, Max Schrems argued that:
It is hard to see how this new body would fulfill the formal requirements of a court or tribunal under Article 47 CFR, especially when compared to ongoing cases and standards applied within the EU (for example in Poland and Hungary).
This comment raises two distinct issues. First, Schrems seems to suggest that an adequacy decision can only be granted if the available redress mechanism satisfies the requirements of Article 47 of the Charter. But this is a hasty conclusion. The CJEU’s phrasing in Schrems II is more cautious:
…Article 47 of the Charter, which also contributes to the required level of protection in the European Union, compliance with which must be determined by the Commission before it adopts an adequacy decision pursuant to Article 45(1) of the GDPR
In arguing that Article 47 “also contributes to the required level of protection,” the Court is not saying that it determines the required level of protection. This is potentially significant, given that the standard of adequacy is “essential equivalence,” not that it be procedurally and substantively identical. Moreover, the Court did not say that the Commission must determine compliance with Article 47 itself, but with the “required level of protection” (which, again, must be “essentially equivalent”).
Second, there is the related but distinct question of whether the redress mechanism is effective under the applicable standard of “required level of protection.” Christakis, Propp, and Swire offered a helpful analysis suggesting that it is, considering the proposed DPRC’s independence, effective investigative powers, and authority to issue binding determinations. I will offer a more detailed analysis of this point in future posts.
Finally, NOYB raised a concern that “judgment by ‘Court’ [is] already spelled out in Executive Order.” This concern seems to be based on the view that a decision of the DPRC (“the judgment”) and what the DPRC communicates to the complainant are the same thing. Or in other words, that legal effects of a DPRC decision are exhausted by providing the individual with the neither-confirm-nor-deny statement set out in Section 3 of the executive order. This is clearly incorrect: the DPRC has power to issue binding directions to intelligence agencies. The actual binding determinations of the DPRC are not predetermined by the executive order, only the information to be provided to the complainant is.
What may call for closer consideration are issues of access to information and data. For example, in La Quadrature du Net, the CJEU looked at the difficult problem of notification of persons whose data has been subject to state surveillance, requiring individual notification “only to the extent that and as soon as it is no longer liable to jeopardise” the law-enforcement tasks in question. Given the “essential equivalence” standard applicable to third-country adequacy assessments, however, it does not automatically follow that individual notification is required in that context.
Moreover, it also does not necessarily follow that adequacy requires that EU citizens have a right to access the data processed by foreign government agencies. The fact that there are significant restrictions on rights to information and to access in some EU member states, though not definitive (after all, those countries may be violating EU law), may be instructive for the purposes of assessing the adequacy of data protection in a third country, where EU law requires only “essential equivalence.”
Conclusion
There are difficult questions of EU law that the European Commission will need to address in the process of deciding whether to issue a new adequacy decision for the United States. It is also clear that an affirmative decision from the Commission will be challenged before the CJEU, although the arguments for such a challenge are not yet well-developed. In future posts I will provide more detailed analysis of the pivotal legal questions. My focus will be to engage with the forthcoming legal analyses from Schrems and NOYB and from other careful observers.
The concept of European “digital sovereignty” has been promoted in recent years both by high officials of the European Union and by EU national governments. Indeed, France made strengthening sovereignty one of the goals of its recent presidency in the EU Council.
The approach taken thus far both by the EU and by national authorities has been not to exclude foreign businesses, but instead to focus on research and development funding for European projects. Unfortunately, there are worrying signs that this more measured approach is beginning to be replaced by ill-conceived moves toward economic protectionism, ostensibly justified by national-security and personal-privacy concerns.
In this context, it is worth reconsidering why Europeans’ best interests are best served not by economic isolationism, but by an understanding of sovereignty that capitalizes on alliances with other free democracies.
Protectionism Under the Guise of Cybersecurity
Among the primary worrying signs regarding the EU’s approach to digital sovereignty is the union’s planned official cybersecurity-certification scheme. The European Commission is reportedly pushing for “digital sovereignty” conditions in the scheme, which would include data and corporate-entity localization and ownership requirements. This can be categorized as “hard” data localization in the taxonomy laid out by Peter Swire and DeBrae Kennedy-Mayo of Georgia Institute of Technology, in that it would prohibit both data transfers to other countries and for foreign capital to be involved in processing even data that is not transferred.
The European Cybersecurity Certification Scheme for Cloud Services (EUCS) is being prepared by ENISA, the EU cybersecurity agency. The scheme is supposed to be voluntary at first, but it is expected that it will become mandatory in the future, at least for some situations (e.g., public procurement). It was not initially billed as an industrial-policy measure and was instead meant to focus on technical security issues. Moreover, ENISA reportedly did not see the need to include such “digital sovereignty” requirements in the certification scheme, perhaps because they saw them as insufficiently grounded in genuine cybersecurity needs.
Despite ENISA’s position, the European Commission asked the agency to include the digital–sovereignty requirements. This move has been supported by a coalition of European businesses that hope to benefit from the protectionist nature of the scheme. Somewhat ironically, their official statement called on the European Commission to “not give in to the pressure of the ones who tend to promote their own economic interests,”
The governments of Denmark, Estonia, Greece, Ireland, Netherlands, Poland, and Sweden expressed “strong concerns” about the Commission’s move. In contrast, Germany called for a political discussion of the certification scheme that would take into account “the economic policy perspective.” In other words, German officials want the EU to consider using the cybersecurity-certification scheme to achieve protectionist goals.
Cybersecurity certification is not the only avenue by which Brussels appears to be pursuing protectionist policies under the guise of cybersecurity concerns. As highlighted in a recent report from the Information Technology & Innovation Foundation, the European Commission and other EU bodies have also been downgrading or excluding U.S.-owned firms from technical standard-setting processes.
Do Security and Privacy Require Protectionism?
As others have discussed at length (in addition to Swire and Kennedy-Mayo, also Theodore Christakis) the evidence for cybersecurity and national-security arguments for hard data localization have been, at best, inconclusive. Press reports suggest that ENISA reached a similar conclusion. There may be security reasons to insist upon certain ways of distributing data storage (e.g., across different data centers), but those reasons are not directly related to the division of national borders.
In fact, as illustrated by the well-known architectural goal behind the design of the U.S. military computer network that was the precursor to the Internet, security is enhanced by redundant distribution of data and network connections in a geographically dispersed way. The perils of putting “all one’s data eggs” in one basket (one locale, one data center) were amply illustrated when a fire in a data center of a French cloud provider, OVH, famously brought down millions of websites that were only hosted there. (Notably, OVH is among the most vocal European proponents of hard data localization).
Moreover, security concerns are clearly not nearly as serious when data is processed by our allies as it when processed by entities associated with less friendly powers. Whatever concerns there may be about U.S. intelligence collection, it would be detached from reality to suggest that the United States poses a national-security risk to EU countries. This has become even clearer since the beginning of the Russian invasion of Ukraine. Indeed, the strength of the U.S.-EU security relationship has been repeatedly acknowledged by EU and national officials.
Another commonly used justification for data localization is that it is required to protect Europeans’ privacy. The radical version of this position, seemingly increasingly popular among EU data-protection authorities, amounts to a call to block data flows between the EU and the United States. (Most bizarrely, Russia seems to receive a more favorable treatment from some European bureaucrats). The legal argument behind this view is that the United States doesn’t have sufficient legal safeguards when its officials process the data of foreigners.
The soundness of that view is debated, but what is perhaps more interesting is that similar privacy concerns have also been identified by EU courts with respect to several EU countries. The reaction of those European countries was either to ignore the courts, or to be “ruthless in exploiting loopholes” in court rulings. It is thus difficult to treat seriously the claims that Europeans’ data is much better safeguarded in their home countries than if it flows in the networks of the EU’s democratic allies, like the United States.
Digital Sovereignty as Industrial Policy
Given the above, the privacy and security arguments are unlikely to be the real decisive factors behind the EU’s push for a more protectionist approach to digital sovereignty, as in the case of cybersecurity certification. In her 2020 State of the Union speech, EU Commission President Ursula von der Leyen stated that Europe “must now lead the way on digital—or it will have to follow the way of others, who are setting these standards for us.”
She continued: “On personalized data—business to consumer—Europe has been too slow and is now dependent on others. This cannot happen with industrial data.” This framing suggests an industrial-policy aim behind the digital-sovereignty agenda. But even in considering Europe’s best interests through the lens of industrial policy, there are reasons to question the manner in which “leading the way on digital” is being implemented.
Limitations on foreign investment in European tech businesses come with significant costs to the European tech ecosystem. Those costs are particularly high in the case of blocking or disincentivizing American investment.
Effect on startups
Early-stage investors such as venture capitalists bring more than just financial capital. They offer expertise and other vital tools to help the businesses in which they invest. It is thus not surprising that, among the best investors, those with significant experience in a given area are well-represented. Due to the successes of the U.S. tech industry, American investors are especially well-positioned to play this role.
In contrast, European investors may lack the needed knowledge and skills. For example, in its report on building “deep tech” companies in Europe, Boston Consulting Group noted that a “substantial majority of executives at deep-tech companies and more than three-quarters of the investors we surveyed believe that European investors do not have a good understanding of what deep tech is.”
More to the point, even where EU players do hold advantages, a cooperative economic and technological system will allow the comparative advantage of both U.S. and EU markets to redound to each others’ benefit. That is to say, of course not all U.S. investment expertise will apply in the EU, but certainly some will. Similarly, there will be EU firms that are positioned to share their expertise in the United States. But there is no ex ante way to know when and where these complementarities will exist, which essentially dooms efforts at centrally planning technological cooperation.
Given the close economic, cultural, and historical ties of the two regions, it makes sense to work together, particularly given the rising international-relations tensions outside of the western sphere. It also makes sense, insofar as the relatively open private-capital-investment environment in the United States is nearly impossible to match, let alone surpass, through government spending.
For example, national government and EU funding in Europe has thus far ranged from expensive failures (the “Google-killer”) to the all-too-predictable bureaucracy-heavy grantmaking, the beneficiaries of which describe as lacking flexibility, “slow,” “heavily process-oriented,” and expensive for businesses to navigate. As reported by the Financial Times’ Sifted website, the EU’s own startup-investment scheme (the European Innovation Council) backed only one business over more than a year, and it had “delays in payment” that “left many startups short of cash—and some on the brink of going out of business.”
Starting new business ventures is risky, especially for the founders. They risk devoting their time, resources, and reputation to an enterprise that may very well fail. Given this risk of failure, the potential upside needs to be sufficiently high to incentivize founders and early employees to take the gamble. This upside is normally provided by the possibility of selling one’s shares in a business. In BCG’s previously cited report on deep tech in Europe, respondents noted that the European ecosystem lacks “clear exit opportunities”:
Some investors fear being constrained by European sovereignty concerns through vetoes at the state or Europe level or by rules potentially requiring European ownership for deep-tech companies pursuing strategically important technologies. M&A in Europe does not serve as the active off-ramp it provides in the US. From a macroeconomic standpoint, in the current environment, investment and exit valuations may be impaired by inflation or geopolitical tensions.
More broadly, those exit opportunities also factor importantly into funders’ appetite to price the risk of failure in their ventures. Where the upside is sufficiently large, an investor might be willing to experiment in riskier ventures and be suitably motivated to structure investments to deal with such risks. But where the exit opportunities are diminished, it makes much more sense to spend time on safer bets that may provide lower returns, but are less likely to fail. Coupled with the fact that government funding must run through bureaucratic channels, which are inherently risk averse, the overall effect is a less dynamic funding system.
The Central and Eastern Europe (CEE) region is an especially good example of the positive influence of American investment in Europe’s tech ecosystem. According to the state-owned Polish Development Fund and Dealroom.co, in 2019, $0.9 billion of venture-capital investment in CEE came from the United States, $0.5 billion from Europe, and $0.1 billion from the rest of the world.
Direct investment
Technological investment is rarely, if ever, a zero-sum game. U.S. firms that invest in the EU (and vice versa) do not do so as foreign conquerors, but as partners whose own fortunes are intertwined with their host country. Consider, for example, Google’s recent PLN 2.7 billion investment in Poland. Far from extractive, that investment will build infrastructure in Poland, and will employ an additional 2,500 Poles in the company’s cloud-computing division. This sort of partnership plants the seeds that grow into a native tech ecosystem. The Poles that today work in Google’s cloud-computing division are the founders of tomorrow’s innovative startups rooted in Poland.
The funding that accompanies native operations of foreign firms also has a direct impact on local economies and tech ecosystems. More local investment in technology creates demand for education and support roles around that investment. This creates a virtuous circle that ultimately facilitates growth in the local ecosystem. And while this direct investment is important for large countries, in smaller countries, it can be a critical component in stimulating their own participation in the innovation economy.
According to Crunchbase, out of 2,617 EU-headquartered startups founded since 2010 with total equity funding amount of at least $10 million, 927 (35%) had at least one founder who previously worked for an American company. For example, two of the three founders of Madrid-based Seedtag (total funding of more than $300 million) worked at Google immediately before starting Seedtag.
It is more difficult to quantify how many early employees of European startups built their experience in American-owned companies, but it is likely to be significant and to become even more so, especially in regions—like Central and Eastern Europe—with significant direct U.S. investment in local talent.
Conclusion
Explicit industrial policy for protectionist ends is—at least, for the time being—regarded as unwise public policy. But this is not to say that countries do not have valid national interests that can be met through more productive channels. While strong data-localization requirements is ultimately counterproductive, particularly among closely allied nations, countries have a legitimate interest in promoting the growth of the technology sector within their borders.
National investment in R&D can yield fruit, particularly when that investment works in tandem with the private sector (see, e.g., the Bayh-Dole Act in the United States). The bottom line, however, is that any intervention should take care to actually promote the ends it seeks. Strong data-localization policies in the EU will not lead to success of the local tech industry, but it will serve to wall the region off from the kind of investment that can make it thrive.
[TOTM: The following is part of a digital symposium by TOTM guests and authors on Antitrust’s Uncertain Future: Visions of Competition in the New Regulatory Landscape. Information on the authors and the entire series of posts is available here.]
Things are heating up in the antitrust world. There is considerable pressure to pass the American Innovation and Choice Online Act (AICOA) before the congressional recess in August—a short legislative window before members of Congress shift their focus almost entirely to campaigning for the mid-term elections. While it would not be impossible to advance the bill after the August recess, it would be a steep uphill climb.
But whether it passes or not, some of the damage from AICOA may already be done. The bill has moved the antitrust dialogue that will harm innovation and consumers. In this post, I will first explain AICOA’s fundamental flaws. Next, I discuss the negative impact that the legislation is likely to have if passed, even if courts and agencies do not aggressively enforce its provisions. Finally, I show how AICOA has already provided an intellectual victory for the approach articulated in the European Union (EU)’s Digital Markets Act (DMA). It has built momentum for a dystopian regulatory framework to break up and break into U.S. superstar firms designated as “gatekeepers” at the expense of innovation and consumers.
The Unseen of AICOA
AICOA’s drafters argue that, once passed, it will deliver numerous economic benefits. Sen. Amy Klobuchar (D-Minn.)—the bill’s main sponsor—has stated that it will “ensure small businesses and entrepreneurs still have the opportunity to succeed in the digital marketplace. This bill will do just that while also providing consumers with the benefit of greater choice online.”
Section 3 of the bill would provide “business users” of the designated “covered platforms” with a wide range of entitlements. This includes preventing the covered platform from offering any services or products that a business user could provide (the so-called “self-preferencing” prohibition); allowing a business user access to the covered platform’s proprietary data; and an entitlement for business users to have “preferred placement” on a covered platform without having to use any of that platform’s services.
These entitlements would provide non-platform businesses what are effectively claims on the platform’s proprietary assets, notwithstanding the covered platform’s own investments to collect data, create services, and invent products—in short, the platform’s innovative efforts. As such, AICOA is redistributive legislation that creates the conditions for unfair competition in the name of “fair” and “open” competition. It treats the behavior of “covered platforms” differently than identical behavior by their competitors, without considering the deterrent effect such a framework will have on consumers and innovation. Thus, AICOA offers rent-seeking rivals a formidable avenue to reap considerable benefits at the expense of the innovators thanks to the weaponization of antitrust to subvert, not improve, competition.
In mandating that covered platforms make their data and proprietary assets freely available to “business users” and rivals, AICOA undermines the underpinning of free markets to pursue the misguided goal of “open markets.” The inevitable result will be the tragedy of the commons. Absent the covered platforms having the ability to benefit from their entrepreneurial endeavors, the law no longer encourages innovation. As Joseph Schumpeter seminally predicted: “perfect competition implies free entry into every industry … But perfectly free entry into a new field may make it impossible to enter it at all.”
To illustrate, if business users can freely access, say, a special status on the covered platforms’ ancillary services without having to use any of the covered platform’s services (as required under Section 3(a)(5)), then platforms are disincentivized from inventing zero-priced services, since they cannot cross-monetize these services with existing services. Similarly, if, under Section 3(a)(1) of the bill, business users can stop covered platforms from pre-installing or preferencing an app whenever they happen to offer a similar app, then covered platforms will be discouraged from investing in or creating new apps. Thus, the bill would generate a considerable deterrent effect for covered platforms to invest, invent, and innovate.
AICOA’s most detrimental consequences may not be immediately apparent; they could instead manifest in larger and broader downstream impacts that will be difficult to undo. As the 19th century French economist Frederic Bastiat wrote: “a law gives birth not only to an effect but to a series of effects. Of these effects, the first only is immediate; it manifests itself simultaneously with its cause—it is seen. The others unfold in succession—they are not seen it is well for, if they are foreseen … it follows that the bad economist pursues a small present good, which will be followed by a great evil to come, while the true economist pursues a great good to come,—at the risk of a small present evil.”
To paraphrase Bastiat, AICOA offers ill-intentioned rivals a “small present good”–i.e., unconditional access to the platforms’ proprietary assets–while society suffers the loss of a greater good–i.e., incentives to innovate and welfare gains to consumers. The logic is akin to those who advocate the abolition of intellectual-property rights: The immediate (and seen) gain is obvious, concerning the dissemination of innovation and a reduction of the price of innovation, while the subsequent (and unseen) evil remains opaque, as the destruction of the institutional premises for innovation will generate considerable long-term innovation costs.
Fundamentally, AICOA weakens the benefits of scale by pursuing vertical disintegration of the covered platforms to the benefit of short-term static competition. In the long term, however, the bill would dampen dynamic competition, ultimately harming consumer welfare and the capacity for innovation. Themeasure’s opportunity costs will prevent covered platforms’ innovations from benefiting other business users or consumers. They personify the “unseen,” as Bastiat put it: “[they are] always in the shadow, and who, personifying what is not seen, [are] an essential element of the problem. [They make] us understand how absurd it is to see a profit in destruction.”
The costs could well amount to hundreds of billions of dollars for the U.S. economy, even before accounting for the costs of deterred innovation. The unseen is costly, the seen is cheap.
A New Robinson-Patman Act?
Most antitrust laws are terse, vague, and old: The Sherman Act of 1890, the Federal Trade Commission Act, and the Clayton Act of 1914 deal largely in generalities, with considerable deference for courts to elaborate in a common-law tradition on the specificities of what “restraints of trade,” “monopolization,” or “unfair methods of competition” mean.
In 1936, Congress passed the Robinson-Patman Act, designed to protect competitors from the then-disruptive competition of large firms who—thanks to scale and practices such as price differentiation—upended traditional incumbents to the benefit of consumers. Passed after “Congress made no factual investigation of its own, and ignored evidence that conflicted with accepted rhetoric,” the law prohibits price differentials that would benefit buyers, and ultimately consumers, in the name of less vigorous competition from more efficient, more productive firms. Indeed, under the Robinson-Patman Act, manufacturers cannot give a bigger discount to a distributor who would pass these savings onto consumers, even if the distributor performs extra services relative to others.
Former President Gerald Ford declared in 1975 that the Robinson-Patman Act “is a leading example of [a law] which restrain[s] competition and den[ies] buyers’ substantial savings…It discourages both large and small firms from cutting prices, making it harder for them to expand into new markets and pass on to customers the cost-savings on large orders.” Despite this, calls to amend or repeal the Robinson-Patman Act—supported by, among others, competition scholars like Herbert Hovenkamp and Robert Bork—have failed.
In the 1983 Abbott decision, Justice Lewis Powell wrote: “The Robinson-Patman Act has been widely criticized, both for its effects and for the policies that it seeks to promote. Although Congress is aware of these criticisms, the Act has remained in effect for almost half a century.”
Nonetheless, the act’s enforcement dwindled, thanks to wise reactions from antitrust agencies and the courts. While it is seldom enforced today, the act continues to create considerable legal uncertainty, as it raises regulatory risks for companies who engage in behavior that may conflict with its provisions. Indeed, many of the same so-called “neo-Brandeisians” who support passage of AICOA also advocate reinvigorating Robinson-Patman. More specifically, the new FTC majority has expressed that it is eager to revitalizeRobinson-Patman, even as the law protects less efficient competitors. In other words, the Robinson-Patman Act is a zombie law: dead, but still moving.
Even if the antitrust agencies and courts ultimately follow the same path of regulatory and judicial restraint on AICOA that they have on Robinson-Patman, the legal uncertainty its existence will engender will act as a powerful deterrent on disruptive competition that dynamically benefits consumers and innovation. In short, like the Robinson-Patman Act, antitrust agencies and courts will either enforce AICOA–thus, generating the law’s adverse effects on consumers and innovation–or they will refrain from enforcing AICOA–but then, the legal uncertainty shall lead to unseen, harmful effects on innovation and consumers.
For instance, the bill’s prohibition on “self-preferencing” in Section 3(a)(1) will prevent covered platforms from offering consumers new products and services that happen to compete with incumbents’ products and services. Self-preferencing often is a pro-competitive, pro-efficiency practice that companies widely adopt—a reality that AICOA seems to ignore.
Would AICOA prevent, e.g., Apple from offering a bundled subscription to Apple One, which includes Apple Music, so that the company can effectively compete with incumbents like Spotify? As with Robinson-Patman, antitrust agencies and courts will have to choose whether to enforce a productivity-decreasing law, or to ignore congressional intent but, in the process, generate significant legal uncertainties.
Judge Bork once wrote that Robinson-Patman was “antitrust’s least glorious hour” because, rather than improving competition and innovation, it reduced competition from firms who happen to be more productive, innovative, and efficient than their rivals. The law infamously protected inefficient competitors rather than competition. But from the perspective of legislative history perspective, AICOA may be antitrust’s new “least glorious hour.” If adopted, it will adversely affect innovation and consumers, as opportunistic rivals will be able to prevent cost-saving practices by the covered platforms.
As with Robinson-Patman, calls to amend or repeal AICOA may follow its passage. But Robinson-Patman Act illustrates the path dependency of bad antitrust laws. However costly and damaging, AICOA would likely stay in place, with regular calls for either stronger or weaker enforcement, depending on whether the momentum shifts from populist antitrust or antitrust more consistent with dynamic competition.
Victory of the Brussels Effect
The future of AICOA does not bode well for markets, either from a historical perspective or from a comparative-law perspective. The EU’s DMA similarly targets a few large tech platforms but it isbroader, harsher, and swifter. In the competition between these two examples of self-inflicted techlash, AICOA will pale in comparison with the DMA. Covered platforms will be forced to align with theDMA’s obligations and prohibitions.
Consequently, AICOA is a victory of the DMA and of the Brussels effect in general. AICOA effectively crowns the DMA as the all-encompassing regulatory assault on digital gatekeepers. While members of Congress have introduced numerous antitrust bills aimed at targeting gatekeepers, the DMA is the one-stop-shop regulation that encompasses multiple antitrust bills and imposes broader prohibitions and stronger obligations on gatekeepers. In other words, the DMA outcompetes AICOA.
Commentators seldom lament the extraterritorial impact of European regulations. Regarding regulating digital gatekeepers, U.S. officials should have pushed back against the innovation-stifling, welfare-decreasing effects of the DMA on U.S. tech companies, in particular, and on U.S. technological innovation, in general. To be fair, a few U.S. officials, such as Commerce Secretary Gina Raimundo, didvoice opposition to the DMA. Indeed, well-aware of the DMA’s protectionist intent and its potential tobreak up and break into tech platforms, Raimundoexpressed concerns that antitrust should not be about protecting competitors and deterring innovation but rather about protecting the process of competition, however disruptive may be.
The influential neo-Brandeisians and radical antitrust reformers, however,lashed out at Raimundo andeffectively shamed the Biden administration into embracing the DMA (and its sister regulation, AICOA). Brussels did not have to exert its regulatory overreach; the U.S. administration happily imports and emulates European overregulation. There is no better way for European officials to see their dreams come true: a techlash against U.S. digital platforms that enjoys the support of local officials.
In that regard, AICOA has already played a significant role in shaping the intellectual mood in Washington and in altering the course of U.S. antitrust. Members of Congress designed AICOA along the linespioneered by the DMA. Sen. Klobuchar has argued that America should emulate European competition policy regarding tech platforms. Lina Khan, now chair of the FTC, co-authoredthe U.S. House Antitrust Subcommittee report, which recommended adopting the European concept of “abuse of dominant position” in U.S. antitrust. In her current position, Khan nowpraises the DMA. Tim Wu, competition counsel for the White House,has praised European competition policy and officials. Indeed, the neo-Brandeisians’ have not only praised the European Commission’s fines against U.S. tech platforms (despiteearly criticisms from former President Barack Obama) but have more dramatically called for the United States to imitate the European regulatory framework.
In this regulatory race to inefficiency, the standard is set in Brussels with the blessings of U.S. officials. Not even the precedent set by the EU’s General Data Protection Regulation (GDPR) fully captures the effects the DMA will have. Privacy laws passed by U.S. states’ privacy have mostly reacted to the reality of the GDPR. With AICOA, Congress is proactively anticipating, emulating, and welcoming the DMA before it has even been adopted. The intellectual and policy shift is historical, and so is the policy error.
AICOA and the Boulevard of Broken Dreams
AICOA is a failure similar to the Robinson-Patman Act and a victory for the Brussels effect and the DMA. Consumers will be the collateral damages, and the unseen effects on innovation will take years before they materialize. Calls for amendments and repeals of AICOA are likely to fail, so that the inevitable costs will forever bear upon consumers and innovation dynamics.
AICOA illustrates the neo-Brandeisian opposition to large innovative companies. Joseph Schumpeter warned against such hostility and its effect on disincentivizing entrepreneurs to innovate when he wrote:
Faced by the increasing hostility of the environment and by the legislative, administrative, and judicial practice born of that hostility, entrepreneurs and capitalists—in fact the whole stratum that accepts the bourgeois scheme of life—will eventually cease to function. Their standard aims are rapidly becoming unattainable, their efforts futile.
President William Howard Taft once said, “the world is not going to be saved by legislation.” AICOA will not save antitrust, nor will consumers. To paraphrase Schumpeter, the bill’s drafters “walked into our future as we walked into the war, blindfolded.” AICOA’s intentions to deliver greater competition, a fairer marketplace, greater consumer choice, and more consumer benefits will ultimately scatter across the boulevard of broken dreams.
The Baron de Montesquieu once wrote that legislators should only change laws with a “trembling hand”:
It is sometimes necessary to change certain laws. But the case is rare, and when it happens, they should be touched only with a trembling hand: such solemnities should be observed, and such precautions are taken that the people will naturally conclude that the laws are indeed sacred since it takes so many formalities to abrogate them.
AICOA’s drafters had a clumsy hand, coupled with what Friedrich Hayek would call “a pretense of knowledge.” They were certain to do social good and incapable of thinking of doing social harm. The future will remember AICOA as the new antitrust’s least glorious hour, where consumers and innovation were sacrificed on the altar of a revitalized populist view of antitrust.
Having earlier passed through subcommittee, the American Data Privacy and Protection Act (ADPPA) has now been cleared for floor consideration by the U.S. House Energy and Commerce Committee. Before the markup, we noted that the ADPPA mimics some of the worst flaws found in the European Union’s General Data Protection Regulation (GDPR), while creating new problems that the GDPR had avoided. Alas, the amended version of the legislation approved by the committee not only failed to correct those flaws, but in some cases it actually undid some of the welcome corrections that had been made to made to the original discussion draft.
Is Targeted Advertising ‘Strictly Necessary’?
The ADPPA’s original discussion draft classified “information identifying an individual’s online activities over time or across third party websites” in the broader category of “sensitive covered data,” for which a consumer’s expression of affirmative consent (“cookie consent”) would be required to collect or process. Perhaps noticing the questionable utility of such a rule, the bill’s sponsors removed “individual’s online activities” from the definition of “sensitive covered data” in the version of ADPPA that was ultimately introduced.
The manager’s amendment from Energy and Commerce Committee Chairman Frank Pallone (D-N.J.) reverted that change and “individual’s online activities” are once again deemed to be “sensitive covered data.” However, the marked-up version of the ADPPA doesn’t require express consent to collect sensitive covered data. In fact, it seems not to consider the possibility of user consent; firms will instead be asked to prove that their collection of sensitive data was a “strict necessity.”
The new rule for sensitive data—in Section 102(2)—is that collecting or processing such data is allowed “where such collection or processing is strictly necessary to provide or maintain a specific product or service requested by the individual to whom the covered data pertains, or is strictly necessary to effect a purpose enumerated” in Section 101(b) (though with exceptions—notably for first-party advertising and targeted advertising).
This raises the question of whether, e.g., the use of targeted advertising based on a user’s online activities is “strictly necessary” to provide or maintain Facebook’s social network? Even if the courts eventually decide, in some cases, that it is necessary, we can expect a good deal of litigation on this point. This litigation risk will impose significant burdens on providers of ad-supported online services. Moreover, it would effectively invite judges to make business decisions, a role for which they are profoundly ill-suited.
Given that the ADPPA includes the “right to opt-out of targeted advertising”—in Section 204(c)) and a special targeted advertising “permissible purpose” in Section 101(b)(17)—this implies that it must be possible for businesses to engage in targeted advertising. And if it is possible, then collecting and processing the information needed for targeted advertising—including information on an “individual’s online activities,” e.g., unique identifiers – Section 2(39)—must be capable of being “strictly necessary to provide or maintain a specific product or service requested by the individual.” (Alternatively, it could have been strictly necessary for one of the other permissible purposes from Section 101(b), but none of them appear to apply to collecting data for the purpose of targeted advertising).
The ADPPA itself thus provides for the possibility of targeted advertising. Therefore, there should be no reason for legal ambiguity about when collecting “individual’s online activities” is “strictly necessary to provide or maintain a specific product or service requested by the individual.” Do we want judges or other government officials to decide which ad-supported services “strictly” require targeted advertising? Choosing business models for private enterprises is hardly an appropriate role for the government. The easiest way out of this conundrum would be simply to revert back to the ill-considered extension of “sensitive covered data” in the ADPPA version that was initially introduced.
Developing New Products and Services
As noted previously, the original ADPPA discussion draft allowed first-party use of personal data to “provide or maintain a specific product or service requested by an individual” (Section 101(a)(1)). What about using the data to develop new products and services? Can a business even request user consent for that? Under the GDPR, that is possible. Under the ADPPA, it may not be.
The general limitation on data use (“provide or maintain a specific product or service requested by an individual”) was retained from the ADPPA original discussion in the version approved by the committee. As originally introduced, the bill included an exception that could have partially addressed the concern in Section 101(b)(2) (emphasis added):
With respect to covered data previously collected in accordance with this Act, notwithstanding this exception, to process such data as necessary to perform system maintenance or diagnostics, to maintain a product or service for which such data was collected, to conduct internal research or analytics, to improve a product or service for which such data was collected …
Arguably, developing new products and services largely involves “internal research or analytics,” which would be covered under this exception. If the business later wanted to invite users of an old service to use a new service, the business could contact them based on a separate exception for first-party marketing and advertising (Section 101(b)(11) of the introduced bill).
This welcome development was reversed in the manager’s amendment. The new text of the exception (now Section 101(b)(2)(C)) is narrower in a key way (emphasis added): “to conduct internal research or analytics to improve a product or service for which such data was collected.” Hence, it still looks like businesses will find it difficult to use first-party data to develop new products or services.
‘De-Identified Data’ Remains Unclear
Our earlier analysis noted significant confusion in the ADPPA’s concept of “de-identified data.” Neither the introduced version nor the markup amendments addressed those concerns, so it seems worthwhile to repeat and update the criticism here. The drafters seemed to be aiming for a partial exemption from the default data-protection regime for datasets that no longer contain personally identifying information, but that are derived from datasets that once did. Instead of providing such an exemption, however, the rules for de-identified data essentially extend the ADPPA’s scope to nonpersonal data, while also creating a whole new set of problems.
The basic problem is that the definition of “de-identified data” in the ADPPA is not limited to data derived from identifiable data. In the marked-up version, the definition covers: “information that does not identify and is not linked or reasonably linkable to a distinct individual or a device, regardless of whether the information is aggregated.” In other words, it is the converse of “covered data” (personal data): whatever is not “covered data” is “de-identified data.” Even if some data are not personally identifiable and are not a result of a transformation of data that was personally identifiable, they still count as “de-identified data.” If this reading is correct, it creates an absurd result that sweeps all information into the scope of the ADPPA.
For the sake of argument, let’s assume that this confusion can be fixed and that the definition of “de-identified data” is limited to data that is:
derived from identifiable data but
that hold a possibility of re-identification (weaker than “reasonably linkable”) and
are processed by the entity that previously processed the original identifiable data.
Remember that we are talking about data that are not “reasonably linkable to an individual.” Hence, the intent appears to be that the rules on de-identified data would apply to nonpersonal data that would otherwise not be covered by the ADPPA.
The rationale for this may be that it is difficult, legally and practically, to differentiate between personally identifiable data and data that are not personally identifiable. A good deal of seemingly “anonymous” data may be linked to an individual—e.g., by connecting the dataset at hand with some other dataset.
The case for regulation in an example where a firm clearly dealt with personal data, and then derived some apparently de-identified data from them, may actually be stronger than in the case of a dataset that was never directly derived from personal data. But is that case sufficient to justify the ADPPA’s proposed rules?
The ADPPA imposes several duties on entities dealing with “de-identified data” in Section 2(12) of the marked-up version:
To take “reasonable technical measures to ensure that the information cannot, at any point, be used to re-identify any individual or device that identifies or is linked or reasonably linkable to an individual”;
To publicly commit “in a clear and conspicuous manner—
to process and transfer the information solely in a de-identified form without any reasonable means for re-identification; and
to not attempt to re-identify the information with any individual or device that identifies or is linked or reasonably linkable to an individual;”
To “contractually obligate[] any person or entity that receives the information from the covered entity or service provider” to comply with all of the same rules and to include such an obligation “in all subsequent instances for which the data may be received.”
The first duty is superfluous and adds interpretative confusion, given that de-identified data, by definition, are not “reasonably linkable” with individuals.
The second duty — public commitment — unreasonably restricts what can be done with nonpersonal data. Firms may have many legitimate reasons to de-identify data and then to re-identify them later. This provision would effectively prohibit firms from attempts at data minimization (resulting in de-identification) if those firms may at any point in the future need to link the data with individuals. It seems that the drafters had some very specific (and likely rare) mischief in mind here, but ended up prohibiting a vast sphere of innocuous activity.
Note that, for data to become “de-identified data,” they must first be collected and processed as “covered data” in conformity with the ADPPA and then transformed (de-identified) in such a way as to no longer meet the definition of “covered data.” If someone then re-identifies the data, this will again constitute “collection” of “covered data” under the ADPPA. At every point of the process, personally identifiable data is covered by the ADPPA rules on “covered data.”
Finally, the third duty—“share alike” (to “contractually obligate[] any person or entity that receives the information from the covered entity to comply”)—faces a very similar problem as the second duty. Under this provision, the only way to preserve the option for a third party to identify the individuals linked to the data will be for the third party to receive the data in a personally identifiable form. In other words, this provision makes it impossible to share data in a de-identified form while preserving the possibility of re-identification.
Logically speaking, we would have expected a possibility to share the data in a de-identified form; this would align with the principle of data minimization. What the ADPPA does instead is to effectively impose a duty to share de-identified personal data together with identifying information. This is a truly bizarre result, directly contrary to the principle of data minimization.
Fundamental Issues with Enforcement
One of the most important problems with the ADPPA is its enforcement provisions. Most notably, the private right of action creates pernicious incentives for excessive litigation by providing for both compensatory damages and open-ended injunctive relief. Small businesses have a right to cure before damages can be sought, but many larger firms are not given a similar entitlement. Given such open-ended provisions as whether using web-browsing behavior is “strictly necessary” to improve a product or service, the litigation incentives become obvious. At the very least, there should be a general opportunity to cure, particularly given the broad restrictions placed on essentially all data use.
The bill also creates multiple overlapping power centers for enforcement (as we have previously noted):
The bill carves out numerous categories of state law that would be excluded from pre-emption… as well as several specific state laws that would be explicitly excluded, including Illinois’ Genetic Information Privacy Act and elements of the California Consumer Privacy Act. These broad carve-outs practically ensure that ADPPA will not create a uniform and workable system, and could potentially render the entire pre-emption section a dead letter. As written, it offers the worst of both worlds: a very strict federal baseline that also permits states to experiment with additional data-privacy laws.
Unfortunately, the marked-up version appears to double down on these problems. For example, the bill pre-empts the Federal Communication Commission (FCC) from enforcing sections 222, 338(i), and 631 of the Communications Act, which pertain to privacy and data security. An amendment was offered that would have pre-empted the FCC from enforcing any provisions of the Communications Act (e.g., sections 201 and 202) for data-security and privacy purposes, but it was withdrawn. Keeping two federal regulators on the beat for a single subject area creates an inefficient regime. The FCC should be completely pre-empted from regulating privacy issues for covered entities.
The amended bill also includes an ambiguous provision that appears to serve as a partial carveout for enforcement by the California Privacy Protection Agency (CCPA). Some members of the California delegation—notably, committee members Anna Eshoo and Doris Matsui (both D-Calif.)—have expressed concern that the bill would pre-empt California’s own California Privacy Rights Act. A proposed amendment by Eshoo to clarify that the bill was merely a federal “floor” and that state laws may go beyond ADPPA’s requirements failed in a 48-8 roll call vote. However, the marked-up version of the legislation does explicitly specify that the CPPA “may enforce this Act, in the same manner, it would otherwise enforce the California Consumer Privacy Act.” How courts might interpret this language should the CPPA seek to enforce provisions of the CCPA that otherwise conflict with the ADPPA is unclear, thus magnifying the problem of compliance with multiple regulators.
Conclusion
As originally conceived, the basic conceptual structure of the ADPPA was, to a very significant extent, both confused and confusing. Not much, if anything, has since improved—especially in the marked-up version that regressed the ADPPA to some of the notably bad features of the original discussion draft. The rules on de-identified data are also very puzzling: their effect contradicts the basic principle of data minimization that the ADPPA purports to uphold. Those examples strongly suggest that the ADPPA is still far from being a properly considered candidate for a comprehensive federal privacy legislation.
European Union lawmakers appear closeto finalizing a number of legislative proposals that aim to reform the EU’s financial-regulation framework in response to the rise of cryptocurrencies. Prominent within the package are new anti-money laundering and “countering the financing of terrorism” rules (AML/CFT), including an extension of the so-called “travel rule.” The travel rule, which currently applies to wire transfers managed by global banks, would be extended to require crypto-asset service providers to similarly collect and make available details about the originators and beneficiaries of crypto-asset transfers.
This legislative process proceeded with unusual haste in recent months, which partially explains why legal objections to the proposals have not been adequately addressed. The resulting legislation is fundamentally flawed to such an extent that some of its key features are clearly invalid under EU primary (treaty) law and liable to be struck down by the Court of Justice of the European Union (CJEU).
In this post, I will offer a brief overview of some of the concerns, which I also discuss in this recent Twitter thread. I focus primarily on the travel rule, which—in the light of EU primary law—constitutes a broad and indiscriminate surveillance regime for personal data. This characterization also applies to most of AML/CFT.
The CJEU, the EU’s highest court, established a number of conditions that such legally mandated invasions of privacy must satisfy in order to be valid under EU primary law (the EU Charter of Fundamental Rights). The legal consequences of invalidity are illustrated well by the Digital Rights Ireland judgment, in which the CJEU struck down an entire piece of EU legislation (the Data Retention Directive). Alternatively, the CJEU could decide to interpret EU law as if it complied with primary law, even if that is contrary to the text.
The Travel Rule in the Transfer of Funds Regulation
The EU travel rule is currently contained in the 2015 Wire Transfer Regulation (WTR). But at the end of June, EU legislators reached a likely final deal on its replacement, the Transfer of Funds Regulation (TFR; see the original proposal from July 2021). I focus here on the TFR, but much of the argument also applies to the older WTR now in force.
The TFR imposes obligations on payment-system providers and providers of crypto-asset transfers (refer to here, collectively, as “service providers”) to collect, retain, transfer to other service providers, and—in some cases—report to state authorities:
…information on payers and payees, accompanying transfers of funds, in any currency, and the information on originators and beneficiaries, accompanying transfers of crypto-assets, for the purposes of preventing, detecting and investigating money laundering and terrorist financing, where at least one of the payment or crypto-asset service providers involved in the transfer of funds or crypto-assets is established in the Union. (Article 1 TFR)
The TFR’s scope extends to money transfers between bank accounts or other payment accounts, as well as transfers of crypto assets other than peer-to-peer transfers without the involvement of a service provider (Article 2 TFR). Hence, the scope of the TFR includes, but is not limited to, all those who send or receive bank transfers. This constitutes the vast majority of adult EU residents.
The information that service providers are obligated to collect and retain (under Articles 4, 10, 14, and 21 TFR) include data that allow for the identification of both sides of a transfer of funds (the parties’ names, as well as the address, country, official personal document number, customer identification number, or the sender’s date and place of birth) and for linking their identity with the (payment or crypto-asset) account number or crypto-asset wallet address. The TFR also obligates service providers to collect and retain additional data to verify the accuracy of the identifying information “on the basis of documents, data or information obtained from a reliable and independent source” (Articles 4(4), 7(3), 14(5), 16(2) TFR).
The scope of the obligation to collect and retain verification data is vague and is likely to lead service providers to require their customers to provide copies of passports, national ID documents, bank or payment-account statements, and utility bills, as is the case under the WTR and the 5th AML Directive. Such data is overwhelmingly likely to go beyond information on the civil identity of customers and will often, if not almost always, allow inferring even sensitive personal data about the customer.
The data-collection and retention obligations in the TFR are general and indiscriminate. No distinction is made in TFR’s data-collection and retention provisions based on likelihood of a connection with criminal activity, except for verification data in the case of transfers of funds (an exception not applicable to crypto assets). Even, the distinction in the case of verification data for transfers of funds (“has reasonable grounds for suspecting money laundering or terrorist financing”) arguably lacks the precision required under CJEU case law.
Analogies with the CJEU’s Passenger Name Records Decision
In late June, following its established approach in similar cases, the CJEU gave its judgment in the Ligue des droits humainscase, which challenged the EU and Belgian regimes on passenger name records (PNR). The CJEU decided there that the applicable EU law, the PNR Directive, is valid under EU primary law. But it reached that result by interpreting some of the directive’s provisions in ways contrary to their express language and by deciding that some national legal rules implementing the directive are invalid. Some features of the PNR regime that were challenged by the court are strikingly similar to the TFR regime.
First, just like the TFR, the PNR rules imposed a five-year data-retention period for the data of all passengers, even where there is no “objective evidence capable of establishing a risk that relates to terrorist offences or serious crime having an objective link, even if only an indirect one, with those passengers’ air travel.” The court decided that this was a disproportionate restriction of the rights to privacy and to the protection of personal data under Articles 5-7 of the EU Charter of Fundamental Rights. Instead of invalidating the relevant article of the PNR Directive, the CJEU reinterpreted it as if it only allowed for five-year retention in cases where there is evidence of a relevant connection to criminality.
Applying analogous reasoning to the TFR, which imposes an indiscriminate five-year data retention period in its Article 21, the conclusion must be that this TFR provision is invalid under Articles 7-8 of the charter. Article 21 TFR may, at minimum, need to be recast to apply only to that transaction data where there is “objective evidence capable of establishing a risk” that it is connected to serious crime. The court also considered the issue of government access to data that has already been collected. Under the CJEU’s established interpretation of the EU Charter, “it is essential that access to retained data by the competent authorities be subject to a prior review carried out either by a court or by an independent administrative body.” In the PNR regime, at least some countries (such as Belgium) assigned this role to their “passenger information units” (PIUs). The court noted that a PIU is “an authority competent for the prevention, detection, investigation and prosecution of terrorist offences and of serious crime, and that its staff members may be agents seconded from the competent authorities” (e.g. from police or intelligence authorities). But according to the court:
That requirement of independence means that that authority must be a third party in relation to the authority which requests access to the data, in order that the former is able to carry out the review, free from any external influence. In particular, in the criminal field, the requirement of independence entails that the said authority, first, should not be involved in the conduct of the criminal investigation in question and, secondly, must have a neutral stance vis-a-vis the parties to the criminal proceedings …
The CJEU decided that PIUs do not satisfy this requirement of independence and, as such, cannot decide on government access to the retained data.
The TFR (especially its Article 19 on provision of information) does not provide for prior independent review of access to retained data. To the extent that such a review is conducted by Financial Intelligence Units (FIUs) under the AML Directive, concerns arise very similar to the treatment of PIUs under the PNR regime. While Article 32 of the AML Directive requires FIUs to be independent, that doesn’t necessarily mean that they are independent in the ways required of the authority that will decide access to retained data under Articles 7-8 of the EU Charter. For example, the AML Directive does not preclude the possibility of seconding public prosecutors, police, or intelligence officers to FIUs.
It is worth noting that none of the conclusions reached by the CJEU in the PNR case are novel; they are well-grounded in established precedent.
A General Proportionality Argument
Setting aside specific analogies with previous cases, the TFR clearly has not been accompanied by a more general and fundamental reflection on the proportionality of its basic scheme in the light of the EU Charter. A pressing question is whether the TFR’s far-reaching restrictions of the rights established in Articles 7-8 of the EU Charter (and perhaps other rights, like freedom of expression in Article 11) are strictly necessary and proportionate.
Arguably, the AML/CFT regime—including the travel rule—are significantly more costly and more rights-restricting than potential alternatives. The basic problem is that there is no reliable data on the relative effectiveness of measures like the travel rule. Defenders of the current AML/CFT regime focus on evidence that it contributes to preventing or prosecuting some crime. But this is not the relevant question when it comes to proportionality. The relevant question is whether those measures are as effective or more effective than alternative, less costly, and more privacy-preserving alternatives. One conservative estimate holds that AML compliance costs in Europe were “120 times the amount successfully recovered from criminals’ and exceeded the estimated total of criminal funds (including funds not seized or identified).”
The fact that the current AML/CFT regime is a de facto global standard cannot serve as a sufficient justification either, given that EU fundamental law is perfectly comfortable in rejecting non-European law-enforcement practices (see the CJEU’s decision in Schrems). The travel rule has been unquestioningly imported to EU law from U.S. law (via FATF), where the standards of constitutional protection of privacy are much different than under the EU Charter. This fact would likely be noticed by the Court of Justice in any putative challenge to the TFR or other elements of the AML/CFT regime.
Here, I only flag the possibility of a general proportionality challenge. Much more work needs to be done to flesh it out.
Conclusion
Due to the political and resource constraints of the EU legislative process, it is possible that the legislative proposals in the financial-regulation package did not receive sufficient legal scrutiny from the perspective of their compatibility with the EU Charter of Fundamental Rights. This hypothesis would explain the presence of seemingly clear violations, such as the indiscriminate five-year data-retention period. Given that none of the proposals has, as yet, been voted into law, making the legislators aware of the problem may help to address at least some of the issues.
Legal arguments about the AML/CFT regime’s incompatibility with the EU Charter should be accompanied with concrete alternative proposals to achieve the goals of preventing and combating serious crime that, according to the best evidence, the current AML/CFT regime does ineffectively. We need more regulatory imagination. For example, one part of the solution may be to properly staff and equip government agencies tasked with prosecuting financial crime.
But it’s also possible that the proposals, including the TFR, will be adopted broadly without amendment. In that case, the main recourse available to EU citizens (or to any EU government) will be to challenge the legality of the measures before the Court of Justice.
Just three weeks after a draft version of the legislation was unveiled by congressional negotiators, the American Data Privacy and Protection Act (ADPPA) is heading to its first legislative markup, set for tomorrow morning before the U.S. House Energy and Commerce Committee’s Consumer Protection and Commerce Subcommittee.
Though the bill’s legislative future remains uncertain, particularly in the U.S. Senate, it would be appropriate to check how the measure compares with, and could potentially interact with, the comprehensive data-privacy regime promulgated by the European Union’s General Data Protection Regulation (GDPR). A preliminary comparison of the two shows that the ADPPA risks adopting some of the GDPR’s flaws, while adding some entirely new problems.
American ‘cookie law’: welcome to consent hell
A common misconception about the GDPR is that it imposed a requirement for “cookie consent” pop-ups that mar the experience of European users of the Internet. In fact, this requirement comes from a different and much older piece of EU law, the 2002 ePrivacy Directive. In most circumstances, the GDPR itself does not require express consent for cookies or other common and beneficial mechanisms to keep track of user interactions with a website. Website publishers could likely rely on one of two lawful bases for data processing outlined in Article 6 of the GDPR:
data processing is necessary in connection with a contractual relationship with the user, or
“processing is necessary for the purposes of the legitimate interests pursued by the controller or by a third party” (unless overridden by interests of the data subject).
For its part, the ADPPA generally adopts the “contractual necessity” basis for data processing but excludes the option to collect or process “information identifying an individual’s online activities over time or across third party websites.” The ADPPA instead classifies such information as “sensitive covered data.” It’s difficult to see what benefit users would derive from having to click that they “consent” to features that are clearly necessary for the most basic functionality, such as remaining logged in to a site or adding items to an online shopping cart. But the expected result will be many, many more popup consent queries, like those that already bedevil European users.
Using personal data to create new products
Section 101(a)(1) of the ADPPA expressly allows the use of “covered data” (personal data) to “provide or maintain a specific product or service requested by an individual.” But the legislation is murkier when it comes to the permissible uses of covered data to develop new products. This would only clearly be allowed where each data subject concerned could be asked if they “request” the specific future product. By contrast, under the GDPR, it is clear that a firm can ask for user consent to use their data to develop future products.
Moving beyond Section 101, we can look to the “general exceptions” in Section 209 of the ADPPA, specifically the exception in Section 209(a)(2)):
With respect to covered data previously collected in accordance with this Act, notwithstanding this exception, to perform system maintenance, diagnostics, maintain a product or service for which such covered data was collected, conduct internal research or analytics to improve products and services, perform inventory management or network management, or debug or repair errors that impair the functionality of a service or product for which such covered data was collected by the covered entity, except such data shall not be transferred.
While this provision mentions conducting “internal research or analytics to improve products and services,” it also refers to “a product or service for which such covered data was collected.” The concern here is that this could be interpreted as only allowing “research or analytics” in relation to existing products known to the data subject.
The road ends here for personal data that the firm collects itself. Somewhat paradoxically, the firm could more easily make the case for using data obtained from a third party. Under Section 302(b) of the ADPPA, a firm only has to ensure that it is not processing “third party data for a processing purpose inconsistent with the expectations of a reasonable individual.” Such a relatively broad “reasonable expectations” basis is not available for data collected directly by first-party covered entities.
Under the GDPR, aside from the data subject’s consent, the firm also could rely on its own “legitimate interest” as a lawful basis to process user data to develop new products. It is true, however, that due to requirements that the interests of the data controller and the data subject must be appropriately weighed, the “legitimate interest” basis is probably less popular in the EU than alternatives like consent or contractual necessity.
Developing this path in the ADPPA would arguably provide a more sensible basis for data uses like the reuse of data for new product development. This could be superior even to express consent, which faces problems like “consent fatigue.” These are unlikely to be solved by promulgating detailed rules on “affirmative consent,” as proposed in Section 2 of the ADPPA.
Problems with ‘de-identified data’
Another example of significant confusion in the ADPPA’s the basic conceptual scheme is the bill’s notion of “de-identified data.” The drafters seemed to be aiming for a partial exemption from the default data-protection regime for datasets that no longer contain personally identifying information, but that are derived from datasets that once did. Instead of providing such an exemption, however, the rules for de-identified data essentially extend the ADPPA’s scope to nonpersonal data, while also creating a whole new set of problems.
The basic problem is that the definition of “de-identified data” in the ADPPA is not limited to data derived from identifiable data. The definition covers: “information that does not identify and is not linked or reasonably linkable to an individual or a device, regardless of whether the information is aggregated.” In other words, it is the converse of “covered data” (personal data): whatever is not “covered data” is “de-identified data.” Even if some data are not personally identifiable and are not a result of a transformation of data that was personally identifiable, they still count as “de-identified data.” If this reading is correct, it creates an absurd result that sweeps all information into the scope of the ADPPA.
For the sake of argument, let’s assume that this confusion can be fixed and that the definition of “de-identified data” is limited to data that is:
derived from identifiable data, but
that hold a possibility of re-identification (weaker than “reasonably linkable”) and
are processed by the entity that previously processed the original identifiable data.
Remember that we are talking about data that are not “reasonably linkable to an individual.” Hence, the intent appears to be that the rules on de-identified data would apply to non-personal data that would otherwise not be covered by the ADPPA.
The rationale for this may be that it is difficult, legally and practically, to differentiate between personally identifiable data and data that are not personally identifiable. A good deal of seemingly “anonymous” data may be linked to an individual—e.g., by connecting the dataset at hand with some other dataset.
The case for regulation in an example where a firm clearly dealt with personal data, and then derived some apparently de-identified data from them, may actually be stronger than in the case of a dataset that was never directly derived from personal data. But is that case sufficient to justify the ADPPA’s proposed rules?
The ADPPA imposes several duties on entities dealing with “de-identified data” (that is, all data that are not considered “covered” data):
to take “reasonable measures to ensure that the information cannot, at any point, be used to re-identify any individual or device”;
to publicly commit “in a clear and conspicuous manner—
to process and transfer the information solely in a de- identified form without any reasonable means for re- identification; and
to not attempt to re-identify the information with any individual or device;”
to “contractually obligate[] any person or entity that receives the information from the covered entity to comply with all of the” same rules.
The first duty is superfluous and adds interpretative confusion, given that de-identified data, by definition, are not “reasonably linkable” with individuals.
The second duty — public commitment — unreasonably restricts what can be done with nonpersonal data. Firms may have many legitimate reasons to de-identify data and then to re-identify them later. This provision would effectively prohibit firms from effective attempts at data minimization (resulting in de-identification) if those firms may at any point in the future need to link the data with individuals. It seems that the drafters had some very specific (and likely rare) mischief in mind here but ended up prohibiting a vast sphere of innocuous activity.
Note that, for data to become “de-identified data,” they must first be collected and processed as “covered data” in conformity with the ADPPA and then transformed (de-identified) in such a way as to no longer meet the definition of “covered data.” If someone then re-identifies the data, this will again constitute “collection” of “covered data” under the ADPPA. At every point of the process, personally identifiable data is covered by the ADPPA rules on “covered data.”
Finally, the third duty—“share alike” (to “contractually obligate[] any person or entity that receives the information from the covered entity to comply”)—faces a very similar problem as the second duty. Under this provision, the only way to preserve the option for a third party to identify the individuals linked to the data will be for the third party to receive the data in a personally identifiable form. In other words, this provision makes it impossible to share data in a de-identified form while preserving the possibility of re-identification. Logically speaking, we would have expected a possibility to share the data in a de-identified form; this would align with the principle of data minimization. What the ADPPA does instead is effectively to impose a duty to share de-identified personal data together with identifying information. This is a truly bizarre result, directly contrary to the principle of data minimization.
Conclusion
The basic conceptual structure of the legislation that subcommittee members will take up this week is, to a very significant extent, both confused and confusing. Perhaps in tomorrow’s markup, a more open and detailed discussion of what the drafters were trying to achieve could help to improve the scheme, as it seems that some key provisions of the current draft would lead to absurd results (e.g., those directly contrary to the principle of data minimization).
Given that the GDPR is already a well-known point of reference, including for U.S.-based companies and privacy professionals, the ADPPA may do better to re-use the best features of the GDPR’s conceptual structure while cutting its excesses. Re-inventing the wheel by proposing new concepts did not work well in this ADPPA draft.
[The following is a guest post from Andrew Mercado, a research assistant at the Mercatus Center at George Mason University and an adjunct professor and research assistant at George Mason’s Antonin Scalia Law School.]
Barry Schwartz’s seminal work “The Paradox of Choice” has received substantial attention since its publication nearly 20 years ago. In it, Schwartz argued that, faced with an ever-increasing plethora of products to choose from, consumers often feel overwhelmed and seek to limit the number of choices they must make.
In today’s online digital economy, a possible response to this problem is for digital platforms to use consumer data to present consumers with a “manageable” array of choices and thereby simplify their product selection. Appropriate “curation” of product-choice options may substantially benefit consumer welfare, provided that government regulators stay out of the way.
New Research
In a new paper in the American Economic Review, Mark Armstrong and Jidong Zhou—of Oxford and Yale universities, respectively—develop a theoretical framework to understand how companies compete using consumer data. Their findings conclude that there is, in fact, an impact on consumer, producer, and total welfare when different privacy regimes are enacted to change the amount of information a company can use to personalize recommendations.
The authors note that, at least in theory, there is an optimal situation that maximizes total welfare (scenario one). This is when a platform can aggregate information on consumers to such a degree that buyers and sellers are perfectly matched, leading to consumers buying their first-best option. While this can result in marginally higher prices, understandably leading to higher welfare for producers, search and mismatch costs are minimized by the platform, leading to a high level of welfare for consumers.
The highest level of aggregate consumer welfare comes when product differentiation is minimized (scenario two), leading to a high number of substitutes and low prices. This, however, comes with some level of mismatch. Since consumers are not matched with any recommendations, search costs are high and introduce some error. Some consumers may have had a higher level of welfare with an alternative product, but do not feel the negative effects of such mismatch because of the low prices. Therefore, consumer welfare is maximized, but producer welfare is significantly lower.
Finally, the authors suggest a “nearly total welfare” optimal solution in suggesting a “top two-best” scheme (scenario three), whereby consumers are shown their top two best options without explicit ranking. This nearly maximizes total welfare, since consumers are shown the best options for them and, even if the best match isn’t chosen, the second-best match is close in terms of welfare.
Implications
In cases of platform data aggregation and personalization, scenarios one, two, and three can be represented as different privacy regimes.
Scenario one (a personalized-product regime) is akin to unlimited data gathering, whereby platforms can use as much information as is available to perfectly suggest products based on revealed data. From a competition perspective, interfirm competition will tend to decrease under this regime, since product differentiation will be accentuated, and substitutability will be masked. Since one single product will be shown as the “correct” product, the consumer will not want to shift to a different, welfare-inferior product and firms have incentive to produce ever more specialized products for a relatively higher price. Total welfare under this regime is maximized, with producers using their information to garner a relatively large share of economic surplus. Producers are effectively matched with consumers, and all gains from trade are realized.
Scenario two (a data-privacy regime) is one of near-perfect data privacy, whereby the platform is only able to recommend products based on general information, such as sales trends, new products, or product specifications. Under this regime, competition is maximized, since consumers consider a large pool of goods to be close substitutes. Differences in offered products are downplayed, which has the tendency to reduce prices and increase quality, but at the tradeoff of some consumer-product mismatch. For consumers who want a general product and a low price, this is likely the best option, since prices are low, and competition is high. However, for consumers who want the best product match for their personal use case, they will likely undertake search costs, increasing their opportunity cost of product acquisition and tending toward a total cost closer to the cost under a personalized-product regime.
Scenario three (a curated-list regime) represents defined guardrails surrounding the display of information gathered, along the same lines as the personalized-product regime. Platforms remain able to gather as much information as they desire in order to make a personalized recommendation, but they display an array of products that represent the first two (or three to four, with tighter anti-preference rules) best-choice options. These options are displayed without ranking the products, allowing the consumer to choose from a curated list, rather than a single product. The scenario-three regime has two effects on the market:
It will tend to decrease prices through increased competition. Since firms can know only which consumers to target, not which will choose the product, they have to effectively compete with closely related products.
It will likely spur innovation and increase competition from nascent competitors.
From an innovation perspective, firms will have to find better methods to differentiate themselves from the competition, increasing the probability of a consumer acquiring their product. Also, considering nascent competitors, a new product has an increased chance of being picked when ranked sufficiently high to be included on the consumer’s curated list. In contrast, the probability of acquisition under scenario one’s personalized-product regime is low, since the new product must be a better match than other, existing products. Similarly, under scenario two’s data-privacy regime, there is so much product substitutability in the market that the probability of choosing any one new product is low.
Below is a list of how the regimes stack up:
Personalized-Product: Total welfare is maximized, but prices are relatively higher and competition is relatively lower than under a data-privacy regime.
Data-Privacy: Consumer welfare and competition are maximized, and prices are theoretically minimized, but at the cost of product mismatch. Consumers will face search costs that are not reflected in the prices paid.
Curated-List: Consumer welfare is higher and prices are lower than under a personalized-product regime and competition is lower than under a data-privacy regime, but total welfare is nearly optimal when considering innovation and nascent-competitor effects.
Policy in Context
Applying these theoretical findings to fashion administrable policy prescriptions is understandably difficult. A far easier task is to evaluate the welfare effects of actual and proposed government privacy regulations in the economy. In that light, I briefly assess a recently enacted European data-platform privacy regime and U.S. legislative proposals that would restrict data usage under the guise of bans on “self-preferencing.” I then briefly note the beneficial implications of self-preferencing associated with the two theoretical data-usage scenarios (scenarios one and three) described above (scenario two, data privacy, effectively renders self-preferencing ineffective).
GDPR
The European Union’s General Data Protection Regulation (GDPR)—among the most ambitious and all-encompassing data-privacy regimes to date—has significant negative ramifications for economic welfare. This regulation is most like the second scenario, whereby data collection and utilization are seriously restricted.
The GDPR diminishes competition through its restrictions on data collection and sharing, which reduce the competitive pressure platforms face. For platforms to gain a complete profile of a consumer for personalization, they cannot only rely on data collected on their platform. To ensure a level of personalization that effectively reduces search costs for consumers, these platforms must be able to acquire data from a range of sources and aggregate that data to create a complete profile. Restrictions on aggregation are what lead to diminished competition online.
The GDPR grants consumers the right to choose both how their data is collected and how it is distributed. Not only do platforms themselves have obligations to ensure consumers’ wishes are met regarding their privacy, but firms that sell data to the platform are obligated to ensure the platform does not infringe consumers’ privacy through aggregation.
This creates a high regulatory burden for both the platform and the data seller and reduces the incentive to transfer data between firms. Since the data seller can be held liable for actions taken by the platform, this significantly increases the price at which the data seller will transfer the data. By increasing the risk of regulatory malfeasance, the cost of data must now incorporate some risk premium, reducing the demand for outside data.
This has the effect of decreasing the quality of personalization and tilting the scales toward larger platforms, who have more robust data-collection practices and are able to leverage economies of scale to absorb high regulatory-enforcement costs. The quality of personalization is decreased, since the platform has incentive to create a consumption profile based on activity it directly observes without considering behavior occurring outside of the platform. Additionally, those platforms that are already entrenched and have large user bases are better able to manage the regulatory burden of the GDPR. One survey of U.S. companies with more than 500 workers found that 68% planned to spend between $1 and $10 million in upfront costs to prepare for GDPR compliance, a number that will likely pale in comparison to the long-term compliance costs. For nascent competitors, this outlay of capital represents a significant barrier to entry.
Additionally, as previously discussed, consumers derive some benefit from platforms that can accurately recommend products. If this is the case, then large platforms with vast amounts of accumulated, first-party data will be the consumers’ destination of choice. This will tend to reduce the ability for smaller firms to compete, simply because they do not have access to the same scale of data as the large platforms when data cannot be easily transferred between parties.
Self–Preferencing
Claims of anticompetitive behavior by platforms are abundant (e.g., see here and here), and they often focus on the concept of self-preferencing. Self-preferencing refers to when a company uses its economies of scale, scope, or a combination of the two to offer products at a lower price through an in-house brand. In decrying self-preferencing, many commentators and politicians point to an alleged “unfair advantage” in tech platforms’ ability to leverage data and personalization to drive traffic toward their own products.
It is far from clear, however, that this practice reduces consumer welfare. Indeed, numerous commentaries (e.g., see here and here) circulated since the introduction of anti-preferencing bills in the U.S. Congress (House; Senate) have rejected the notion that self-preferencing is anti-competitive or anti-consumer.
There are good reasons to believe that self-preferencing promotes both competition and consumer welfare. Assume that a company that manufactures or contracts for its own, in-house products can offer them at a marginally lower price for the same relative quality. This decrease in price raises consumer welfare. The in-house brand’s entrance into the market represents a potent competitive threat to firms already producing products, who in turn now have incentive to lower their own prices or raise the quality of their own goods (or both) to maintain their consumer base. This creates even more consumer welfare, since all consumers, not just the ones purchasing the in-house goods, are better off from the entrance of an in-house brand.
It therefore follows that the entrance of an in-house brand and self-preferencing in the data-utilizing regimes discussed above has the potential to enhance consumer welfare.
In general, the use of data analysis on the platform can allow for targeted product entrance into certain markets. If the platform believes it can make a product of similar quality for a lower price, then it will enter that market and consumers will be able to choose a comparable product for a lower price. (If the company does not believe it is able to produce such a product, it will not enter the market with an in-house brand, and consumer welfare will stay the same.) Consumer welfare will further rise as firms producing products that compete against the in-house brand will innovate to compete more effectively.
To be sure, under a personalized-product regime (scenario one), platforms may appear to have an incentive to self-preference to the detriment of consumers. If consumers trust the platform to show the greatest welfare-producing product before the emergence of an in-house brand, the platform may use this consumer trust to its advantage and suggest its own, potentially consumer-welfare-inferior product instead of a competitor’s welfare-superior product. In such a case, consumer welfare may decrease in the face of an in-house brand’s entrance.
The extent of any such welfare loss, however, may be ameliorated (or eliminated entirely) by the platform’s concern that an unexpectedly low level of house-brand product quality will diminish its reputation. Such a reputational loss could come about due to consumer disappointment, plus the efforts of platform rivals to highlight the in-house product’s inferiority. As such, the platform might decide to enhance the quality of its “inferior” in-house offering, or refrain from offering an in-house brand at all.
A curated-list regime (scenario three) is unequivocally consumer-welfare beneficial. Under such a regime, consumers will be shown several more options (a “manageable” number intended to minimize consumer-search costs) than under a personalized-product regime. Consumers can actively compare the offerings from different firms to determine the correct product for their individual use. In this case, there is no incentive to self-preference to the detriment of the consumer, as the consumer is able to make value judgements between the in-house brand and the alternatives.
If the in-house brand is significantly lower in price, but also lower in quality, consumers may not see the two as interchangeable and steer away from the in-house brand. The same follows when the in-house brand is higher in both price and quality. The only instance where the in-house brand has a strong chance of success is when the price is lower than and the quality is greater than competing products. This will tend to increase consumer welfare. Additionally, the entrance of consumer-welfare-superior products into a competitive market will encourage competing firms to innovate and lower prices or raise quality, again increasing consumer welfare for all consumers.
Conclusion
What effects do digital platform-data policies have on consumer welfare? As a matter of theory, if providing an increasing number of product choices does not tend to increase consumer welfare, then do reductions in prices or increases in quality? What about precise targeting of personal-product choices? How about curation—the idea that a consumer raises his or her level of certainty by outsourcing decision-making to a platform that chooses a small set of products for the consumer’s consideration at any given moment? Apart from these theoretical questions, is the current U.S. legal treatment of platform data usage doing a generally good job of promoting consumer welfare? Finally, considering this overview, are new government interventions in platform data policy likely to benefit or harm consumers?
Recently published economic research develops theoretical scenarios that demonstrate how digital platform curation of consumer data may facilitate welfare-enhancing consumer-purchase decisions. At least implicitly, this research should give pause to proponents of major new restrictions of platform data usage.
Furthermore, a review of actual and proposed regulatory restrictions underscores the serious welfare harm of government meddling in digital platform-data usage.
After the first four years of GDPR, it is clear that there have been significant negative unintended consequences stemming from omnibus privacy regulation. Competition has decreased, regulatory barriers to entry have increased, and consumers are marginally worse off. Since companies are less able and willing to leverage data in their operations and service offerings—due in large part to the risk of hefty fines—they are less able to curate and personalize services to consumers.
Additionally, anti-preferencing bills in the United States threaten to suppress the proper functioning of platform markets and reduce consumer welfare by making the utilization of data in product-market decisions illegal. More research is needed to determine the aggregate welfare effects of such preferencing on platforms, but all early indications point to the fact that consumers are better off when an in-house brand enters the market and increases competition.
Furthermore, current U.S. government policy, which generally allows platforms to use consumer data freely, is good for consumer welfare. Indeed, the consumer-welfare benefits generated by digital platforms, which depend critically on large volumes of data, are enormous. This is documented in a well-reasoned Harvard Business Review article (by an MIT professor and his student) that utilizes online choice experiments based on digital-survey techniques.
The message is clear. Governments should avoid new regulatory meddling in digital platform consumer-data usage practices. Such meddling would harm consumers and undermine the economy.
[Judge Douglas Ginsburg was invited to respond to the Beesley Lecture given by Andrea Coscelli, chief executive of the U.K. Competition and Markets Authority (CMA). Both the lecture and Judge Ginsburg’s response were broadcast by the BBC on Oct. 28, 2021. The text of Mr. Coscelli’s Beesley lecture is available on the CMA’s website. Judge Ginsburg’s response follows below.]
Thank you, Victoria, for the invitation to respond to Mr. Coscelli and his proposal for a legislatively founded Digital Markets Unit. Mr. Coscelli is one of the most talented, successful, and creative heads a competition agency has ever had. In the case of the DMU [ed., Digital Markets Unit], however, I think he has let hope triumph over experience and prudence. This is often the case with proposals for governmental reform: Indeed, it has a name, the Nirvana Fallacy, which comes from comparing the imperfectly functioning marketplace with the perfectly functioning government agency. Everything we know about the regulation of competition tells us the unintended consequences may dwarf the intended benefits and the result may be a less, not more, competitive economy. The precautionary principle counsels skepticism about such a major and inherently risky intervention.
Mr. Coscelli made a point in passing that highlights the difference in our perspectives: He said the SMS [ed.,strategic market status] merger regime would entail “a more cautious standard of proof.” In our shared Anglo-American legal culture, a more cautious standard of proof means the government would intervene in fewer, not more, market activities; proof beyond a reasonable doubt in criminal cases is a more cautious standard than a mere preponderance of the evidence. I, too, urge caution, but of the traditional kind.
I will highlight five areas of concern with the DMU proposal.
I. Chilling Effects
The DMU’s ability to designate a firm as being of strategic market significance—or SMS—will place a potential cloud over innovative activity in far more sectors than Mr. Coscelli could mention in his lecture. He views the DMU’s reach as limited to a small number of SMS-designated firms; and that may prove true, but there is nothing in the proposal limiting DMU’s reach.
Indeed, the DMU’s authority to regulate digital markets is surely going to be difficult to confine. Almost every major retail activity or consumer-facing firm involves an increasingly significant digital component, particularly after the pandemic forced many more firms online. Deciding which firms the DMU should cover seems easy in theory, but will prove ever more difficult and cumbersome in practice as digital technology continues to evolve. For instance, now that money has gone digital, a bank is little more than a digital platform bringing together lenders (called depositors) and borrowers, much as Amazon brings together buyers and sellers; so, is every bank with market power and an entrenched position to be subject to rules and remedies laid down by the DMU as well as supervision by the bank regulators? Is Aldi in the crosshairs now that it has developed an online retail platform? Match.com, too? In short, the number of SMS firms will likely grow apace in the next few years.
II. SMS Designations Should Not Apply to the Whole Firm
The CMA’s proposal would apply each SMS designation firm-wide, even if the firm has market power in a single line of business. This will inhibit investment in further diversification and put an SMS firm at a competitive disadvantage across all its businesses.
Perhaps company-wide SMS designations could be justified if the unintended costs were balanced by expected benefits to consumers, but this will not likely be the case. First, there is little evidence linking consumer harm to lines of business in which large digital firms do not have market power. On the contrary, despite the discussion of Amazon’s supposed threat to competition, consumers enjoy lower prices from many more retailers because of the competitive pressure Amazon brings to bear upon them.
Second, the benefits Mr. Coscelli expects the economy to reap from faster government enforcement are, at best, a mixed blessing. The proposal, you see, reverses the usual legal norm, instead making interim relief the rule rather than the exception. If a firm appeals its SMS designation, then under the CMA’s proposal, the DMU’s SMS designations and pro-competition interventions, or PCIs, will not be stayed pending appeal, raising the prospect that a firm’s activities could be regulated for a significant period even though it was improperly designated. Even prevailing in the courts may be a Pyrrhic victory because opportunities will have slipped away. Making matters worse, the DMU’s designation of a firm as SMS will likely receive a high degree of judicial deference, so that errors may never be corrected.
III. The DMU Cannot Be Evidence-based Given its Goals and Objectives
The DMU’s stated goal is to “further the interests of consumers and citizens in digital markets by promoting competition and innovation.”[1] DMU’s objectives for developing codes of conduct are: fair trading, open choices, and trust and transparency.[2] Fairness, openness, trust, and transparency are all concepts that are difficult to define and probably impossible to quantify. Therefore, I fear Mr. Coscelli’s aspiration that the DMU will be an evidence-based, tailored, and predictable regime seem unrealistic. The CMA’s idea of “an evidence-based regime” seems destined to rely mostly upon qualitative conjecture about the potential for the code of conduct to set “rules of the game” that encourage fair trading, open choices, trust, and transparency. Even if the DMU commits to considering empirical evidence at every step of its process, these fuzzy, qualitative objectives will allow it to come to virtually any conclusion about how a firm should be regulated.
Implementing those broad goals also throws into relief the inevitable tensions among them. Some potential conflicts between DMU’s objectives for developing codes of conduct are clear from the EU’s experience. For example, one of the things DMU has considered already is stronger protection for personal data. The EU’s experience with the GDPR shows that data protection is costly and, like any costly requirement, tends to advantage incumbents and thereby discourage new entry. In other words, greater data protections may come at the expense of start-ups or other new entrants and the contribution they would otherwise have made to competition, undermining open choices in the name of data transparency.
Another example of tension is clear from the distinction between Apple’s iOS and Google’s Android ecosystems. They take different approaches to the trade-off between data privacy and flexibility in app development. Apple emphasizes consumer privacy at the expense of allowing developers flexibility in their design choices and offers its products at higher prices. Android devices have fewer consumer-data protections but allow app developers greater freedom to design their apps to satisfy users and are offered at lower prices. The case of Epic Games v. Apple put on display the purportedly pro-competitive arguments the DMU could use to justify shutting down Apple’s “walled garden,” whereas the EU’s GDPR would cut against Google’s open ecosystem with limited consumer protections. Apple’s model encourages consumer trust and adoption of a single, transparent model for app development, but Google’s model encourages app developers to choose from a broader array of design and payment options and allows consumers to choose between the options; no matter how the DMU designs its code of conduct, it will be creating winners and losers at the cost of either “open choices” or “trust and transparency.” As experience teaches is always the case, it is simply not possible for an agency with multiple goals to serve them all at the same time. The result is an unreviewable discretion to choose among them ad hoc.
Finally, notice that none of the DMU’s objectives—fair trading, open choices, and trust and transparency—revolves around quantitative evidence; at bottom, these goals are not amenable to the kind of rigor Mr. Coscelli hopes for.
IV. Speed of Proposals
Mr. Coscelli has emphasized the slow pace of competition law matters; while I empathize, surely forcing merging parties to prove a negative and truncating their due process rights is not the answer.
As I mentioned earlier, it seems a more cautious standard of proof to Mr. Coscelli is one in which an SMS firm’s proposal to acquire another firm is presumed, or all but presumed, to be anticompetitive and unlawful. That is, the DMU would block the transaction unless the firms can prove their deal would not be anticompetitive—an extremely difficult task. The most self-serving version of the CMA’s proposal would require it to prove only that the merger poses a “realistic prospect” of lessening competition, which is vague, but may in practice be well below a 50% chance. Proving that the merged entity does not harm competition will still require a predictive forward-looking assessment with inherent uncertainty, but the CMA wants the costs of uncertainty placed upon firms, rather than it. Given the inherent uncertainty in merger analysis, the CMA’s proposal would pose an unprecedented burden of proof on merging parties.
But it is not only merging parties the CMA would deprive of due process; the DMU’s so-called pro-competitive interventions, or PCI, SMS designations, and code-of-conduct requirements generally would not be stayed pending appeal. Further, an SMS firm could overturn the CMA’s designation only if it could overcome substantial deference to the DMU’s fact-finding. It is difficult to discern, then, the difference between agency decisions and final orders.
The DMU would not have to show or even assert an extraordinary need for immediate relief. This is the opposite of current practice in every jurisdiction with which I am familiar. Interim orders should take immediate effect only in exceptional circumstances, when there would otherwise be significant and irreversible harm to consumers, not in the ordinary course of agency decision making.
V. Antitrust Is Not Always the Answer
Although one can hardly disagree with Mr. Coscelli’s premise that the digital economy raises new legal questions and practical challenges, it is far from clear that competition law is the answer to them all. Some commentators of late are proposing to use competition law to solve consumer protection and even labor market problems. Unfortunately, this theme also recurs in Mr. Coscelli’s lecture. He discusses concerns with data privacy and fair and reasonable contract terms, but those have long been the province of consumer protection and contract law; a government does not need to step in and regulate all realms of activity by digital firms and call it competition law. Nor is there reason to confine needed protections of data privacy or fair terms of use to SMS firms.
Competition law remedies are sometimes poorly matched to the problems a government is trying to correct. Mr. Coscelli discusses the possibility of strong interventions, such as forcing the separation of a platform from its participation in retail markets; for example, the DMU could order Amazon to spin off its online business selling and shipping its own brand of products. Such powerful remedies can be a sledgehammer; consider forced data sharing or interoperability to make it easier for new competitors to enter. For example, if Apple’s App Store is required to host all apps submitted to it in the interest of consumer choice, then Apple loses its ability to screen for security, privacy, and other consumer benefits, as its refusal to deal is its only way to prevent participation in its store. Further, it is not clear consumers want Apple’s store to change; indeed, many prefer Apple products because of their enhanced security.
Forced data sharing would also be problematic; the hiQ v. LinkedIn case in the United States should serve as a cautionary tale. The trial court granted a preliminary injunction forcing LinkedIn to allow hiQ to scrape its users’ profiles while the suit was ongoing. LinkedIn ultimately won the suit because it did not have market power, much less a monopoly, in any relevant market. The court concluded each theory of anticompetitive conduct was implausible, but meanwhile LinkedIn had been forced to allow hiQ to scrape its data for an extended period before the final decision. There is no simple mechanism to “unshare” the data now that LinkedIn has prevailed. This type of case could be common under the CMA proposal because the DMU’s orders will go into immediate effect.
There is potentially much redeeming power in the Digital Regulation Co-operation Forum as Mr. Coscelli described it, but I take a different lesson from this admirable attempt to coordinate across agencies: Perhaps it is time to look beyond antitrust to solve problems that are not based upon market power. As the DRCF highlights, there are multiple agencies with overlapping authority in the digital market space. ICO and Ofcom each have authority to take action against a firm that disseminates fake news or false advertisements. Mr. Coscelli says it would be too cumbersome to take down individual bad actors, but, if so, then the solution is to adopt broader consumer protection rules, not apply an ill-fitting set of competition law rules. For example, the U.K. could change its notice-and-takedown rules to subject platforms to strict liability if they host fake news, even without knowledge that they are doing so, or perhaps only if they are negligent in discharging their obligation to police against it.
Alternatively, the government could shrink the amount of time platforms have to take down information; France gives platforms only about an hour to remove harmful information. That sort of solution does not raise the same prospect of broadly chilling market activity, but still addresses one of the concerns Mr. Coscelli raises with digital markets.
In sum, although Mr. Coscelli is of course correct that competition authorities and governments worldwide are considering whether to adopt broad reforms to their competition laws, the case against broadening remains strong. Instead of relying upon the self-corrective potential of markets, which is admittedly sometimes slower than anyone would like, the CMA assumes markets need regulation until firms prove otherwise. Although clearly well-intentioned, the DMU proposal is in too many respects not met to the task of protecting competition in digital markets; at worst, it will inhibit innovation in digital markets to the point of driving startups and other innovators out of the U.K.
[2] Sam Bowman, Sam Dumitriu & Aria Babu, Conflicting Missions:The Risks of the Digital Markets Unit to Competition and Innovation, Int’l Center for L. & Econ., June 2021, at 13.
The European Commission this week published its proposed Artificial Intelligence Regulation, setting out new rules for “artificial intelligence systems” used within the European Union. The regulation—the commission’s attempt to limit pernicious uses of AI without discouraging its adoption in beneficial cases—casts a wide net in defining AI to include essentially any software developed using machine learning. As a result, a host of software may fall under the regulation’s purview.
The regulation categorizes AIs by the kind and extent of risk they may pose to health, safety, and fundamental rights, with the overarching goal to:
Prohibit “unacceptable risk” AIs outright;
Place strict restrictions on “high-risk” AIs;
Place minor restrictions on “limited-risk” AIs;
Create voluntary “codes of conduct” for “minimal-risk” AIs;
Establish a regulatory sandbox regime for AI systems;
Set up a European Artificial Intelligence Board to oversee regulatory implementation; and
Set fines for noncompliance at up to 30 million euros, or 6% of worldwide turnover, whichever is greater.
AIs That Are Prohibited Outright
The regulation prohibits AI that are used to exploit people’s vulnerabilities or that use subliminal techniques to distort behavior in a way likely to cause physical or psychological harm. Also prohibited are AIs used by public authorities to give people a trustworthiness score, if that score would then be used to treat a person unfavorably in a separate context or in a way that is disproportionate. The regulation also bans the use of “real-time” remote biometric identification (such as facial-recognition technology) in public spaces by law enforcement, with exceptions for specific and limited uses, such as searching for a missing child.
The first prohibition raises some interesting questions. The regulation says that an “exploited vulnerability” must relate to age or disability. In its announcement, the commission says this is targeted toward AIs such as toys that might induce a child to engage in dangerous behavior.
The ban on AIs using “subliminal techniques” is more opaque. The regulation doesn’t give a clear definition of what constitutes a “subliminal technique,” other than that it must be something “beyond a person’s consciousness.” Would this include TikTok’s algorithm, which imperceptibly adjusts the videos shown to the user to keep them engaged on the platform? The notion that this might cause harm is not fanciful, but it’s unclear whether the provision would be interpreted to be that expansive, whatever the commission’s intent might be. There is at least a risk that this provision would discourage innovative new uses of AI, causing businesses to err on the side of caution to avoid the huge penalties that breaking the rules would incur.
The prohibition on AIs used for social scoring is limited to public authorities. That leaves space for socially useful expansions of scoring systems, such as consumers using their Uber rating to show a record of previous good behavior to a potential Airbnb host. The ban is clearly oriented toward more expansive and dystopian uses of social credit systems, which some fear may be used to arbitrarily lock people out of society.
The ban on remote biometric identification AI is similarly limited to its use by law enforcement in public spaces. The limited exceptions (preventing an imminent terrorist attack, searching for a missing child, etc.) would be subject to judicial authorization except in cases of emergency, where ex-post authorization can be sought. The prohibition leaves room for private enterprises to innovate, but all non-prohibited uses of remote biometric identification would be subject to the requirements for high-risk AIs.
Restrictions on ‘High-Risk’ AIs
Some AI uses are not prohibited outright, but instead categorized as “high-risk” and subject to strict rules before they can be used or put to market. AI systems considered to be high-risk include those used for:
Safety components for certain types of products;
Remote biometric identification, except those uses that are banned outright;
Safety components in the management and operation of critical infrastructure, such as gas and electricity networks;
Dispatching emergency services;
Educational admissions and assessments;
Employment, workers management, and access to self-employment;
Evaluating credit-worthiness;
Assessing eligibility to receive social security benefits or services;
A range of law-enforcement purposes (e.g., detecting deepfakes or predicting the occurrence of criminal offenses);
Migration, asylum, and border-control management; and
Administration of justice.
While the commission considers these AIs to be those most likely to cause individual or social harm, it may not have appropriately balanced those perceived harms with the onerous regulatory burdens placed upon their use.
As Mikołaj Barczentewicz at the Surrey Law and Technology Hub has pointed out, the regulation would discourage even simple uses of logic or machine-learning systems in such settings as education or workplaces. This would mean that any workplace that develops machine-learning tools to enhance productivity—through, for example, monitoring or task allocation—would be subject to stringent requirements. These include requirements to have risk-management systems in place, to use only “high quality” datasets, and to allow human oversight of the AI, as well as other requirements around transparency and documentation.
The obligations would apply to any companies or government agencies that develop an AI (or for whom an AI is developed) with a view toward marketing it or putting it into service under their own name. The obligations could even attach to distributors, importers, users, or other third parties if they make a “substantial modification” to the high-risk AI, market it under their own name, or change its intended purpose—all of which could potentially discourage adaptive use.
Without going into unnecessary detail regarding each requirement, some are likely to have competition- and innovation-distorting effects that are worth discussing.
The rule that data used to train, validate, or test a high-risk AI has to be high quality (“relevant, representative, and free of errors”) assumes that perfect, error-free data sets exist, or can easily be detected. Not only is this not necessarily the case, but the requirement could impose an impossible standard on some activities. Given this high bar, high-risk AIs that use data of merely “good” quality could be precluded. It also would cut against the frontiers of research in artificial intelligence, where sometimes only small and lower-quality datasets are available to train AI. A predictable effect is that the rule would benefit large companies that are more likely to have access to large, high-quality datasets, while rules like the GDPR make it difficult for smaller companies to acquire that data.
High-risk AIs also must submit technical and user documentation that detail voluminous information about the AI system, including descriptions of the AI’s elements, its development, monitoring, functioning, and control. These must demonstrate the AI complies with all the requirements for high-risk AIs, in addition to documenting its characteristics, capabilities, and limitations. The requirement to produce vast amounts of information represents another potentially significant compliance cost that will be particularly felt by startups and other small and medium-sized enterprises (SMEs). This could further discourage AI adoption within the EU, as European enterprises already consider liability for potential damages and regulatory obstacles as impediments to AI adoption.
The requirement that the AI be subject to human oversight entails that the AI can be overseen and understood by a human being and that the AI can never override a human user. While it may be important that an AI used in, say, the criminal justice system must be understood by humans, this requirement could inhibit sophisticated uses beyond the reasoning of a human brain, such as how to safely operate a national electricity grid. Providers of high-risk AI systems also must establish a post-market monitoring system to evaluate continuous compliance with the regulation, representing another potentially significant ongoing cost for the use of high-risk AIs.
The regulation also places certain restrictions on “limited-risk” AIs, notably deepfakes and chatbots. Such AIs must be labeled to make a user aware they are looking at or listening to manipulated images, video, or audio. AIs must also be labeled to ensure humans are aware when they are speaking to an artificial intelligence, where this is not already obvious.
Taken together, these regulatory burdens may be greater than the benefits they generate, and could chill innovation and competition. The impact on smaller EU firms, which already are likely to struggle to compete with the American and Chinese tech giants, could prompt them to move outside the European jurisdiction altogether.
Regulatory Support for Innovation and Competition
To reduce the costs of these rules, the regulation also includes a new regulatory “sandbox” scheme. The sandboxes would putatively offer environments to develop and test AIs under the supervision of competent authorities, although exposure to liability would remain for harms caused to third parties and AIs would still have to comply with the requirements of the regulation.
SMEs and startups would have priority access to the regulatory sandboxes, although they must meet the same eligibility conditions as larger competitors. There would also be awareness-raising activities to help SMEs and startups to understand the rules; a “support channel” for SMEs within the national regulator; and adjusted fees for SMEs and startups to establish that their AIs conform with requirements.
These measures are intended to prevent the sort of chilling effect that was seen as a result of the GDPR, which led to a 17% increase in market concentration after it was introduced. But it’s unclear that they would accomplish this goal. (Notably, the GDPR contained similar provisions offering awareness-raising activities and derogations from specific duties for SMEs.) Firms operating in the “sandboxes” would still be exposed to liability, and the only significant difference to market conditions appears to be the “supervision” of competent authorities. It remains to be seen how this arrangement would sufficiently promote innovation as to overcome the burdens placed on AI by the significant new regulatory and compliance costs.
Governance and Enforcement
Each EU member state would be expected to appoint a “national competent authority” to implement and apply the regulation, as well as bodies to ensure high-risk systems conform with rules that require third party-assessments, such as remote biometric identification AIs.
The regulation establishes the European Artificial Intelligence Board to act as the union-wide regulatory body for AI. The board would be responsible for sharing best practices with member states, harmonizing practices among them, and issuing opinions on matters related to implementation.
As mentioned earlier, maximum penalties for marketing or using a prohibited AI (as well as for failing to use high-quality datasets) would be a steep 30 million euros or 6% of worldwide turnover, whichever is greater. Breaking other requirements for high-risk AIs carries maximum penalties of 20 million euros or 4% of worldwide turnover, while maximums of 10 million euros or 2% of worldwide turnover would be imposed for supplying incorrect, incomplete, or misleading information to the nationally appointed regulator.
Is the Commission Overplaying its Hand?
While the regulation only restricts AIs seen as creating risk to society, it defines that risk so broadly and vaguely that benign applications of AI may be included in its scope, intentionally or unintentionally. Moreover, the commission also proposes voluntary codes of conduct that would apply similar requirements to “minimal” risk AIs. These codes—optional for now—may signal the commission’s intent eventually to further broaden the regulation’s scope and application.
The commission clearly hopes it can rely on the “Brussels Effect” to steer the rest of the world toward tighter AI regulation, but it is also possible that other countries will seek to attract AI startups and investment by introducing less stringent regimes.
For the EU itself, more regulation must be balanced against the need to foster AI innovation. Without European tech giants of its own, the commission must be careful not to stifle the SMEs that form the backbone of the European market, particularly if global competitors are able to innovate more freely in the American or Chinese markets. If the commission has got the balance wrong, it may find that AI development simply goes elsewhere, with the EU fighting the battle for the future of AI with one hand tied behind its back.
Policy discussions about the use of personal data often have “less is more” as a background assumption; that data is overconsumed relative to some hypothetical optimal baseline. This overriding skepticism has been the backdrop for sweeping new privacy regulations, such as the California Consumer Privacy Act (CCPA) and the EU’s General Data Protection Regulation (GDPR).
More recently, as part of the broad pushback against data collection by online firms, some have begun to call for creating property rights in consumers’ personal data or for data to be treated as labor. Prominent backers of the idea include New York City mayoral candidate Andrew Yang and computer scientist Jaron Lanier.
The discussion has escaped the halls of academia and made its way into popular media. During a recent discussion with Tesla founder Elon Musk, comedian and podcast host Joe Rogan argued that Facebook is “one gigantic information-gathering business that’s decided to take all of the data that people didn’t know was valuable and sell it and make f***ing billions of dollars.” Musk appeared to agree.
The animosity exhibited toward data collection might come as a surprise to anyone who has taken Econ 101. Goods ideally end up with those who value them most. A firm finding profitable ways to repurpose unwanted scraps is just the efficient reallocation of resources. This applies as much to personal data as to literal trash.
Unfortunately, in the policy sphere, few are willing to recognize the inherent trade-off between the value of privacy, on the one hand, and the value of various goods and services that rely on consumer data, on the other. Ideally, policymakers would look to markets to find the right balance, which they often can. When the transfer of data is hardwired into an underlying transaction, parties have ample room to bargain.
But this is not always possible. In some cases, transaction costs will prevent parties from bargaining over the use of data. The question is whether such situations are so widespread as to justify the creation of data property rights, with all of the allocative inefficiencies they entail. Critics wrongly assume the solution is both to create data property rights and to allocate them to consumers. But there is no evidence to suggest that, at the margin, heightened user privacy necessarily outweighs the social benefits that new data-reliant goods and services would generate. Recent experience in the worlds of personalized medicine and the fight against COVID-19 help to illustrate this point.
Data Property Rights and Personalized Medicine
The world is on the cusp of a revolution in personalized medicine. Advances such as the improved identification of biomarkers, CRISPR genome editing, and machine learning, could usher a new wave of treatments to markedly improve health outcomes.
Personalized medicine uses information about a person’s own genes or proteins to prevent, diagnose, or treat disease. Genetic-testing companies like 23andMe or Family Tree DNA, with the large troves of genetic information they collect, could play a significant role in helping the scientific community to further medical progress in this area.
However, despite the obvious potential of personalized medicine, many of its real-world applications are still very much hypothetical. While governments could act in any number of ways to accelerate the movement’s progress, recent policy debates have instead focused more on whether to create a system of property rights covering personal genetic data.
Some raise concerns that it is pharmaceutical companies, not consumers, who will reap the monetary benefits of the personalized medicine revolution, and that advances are achieved at the expense of consumers’ and patients’ privacy. They contend that data property rights would ensure that patients earn their “fair” share of personalized medicine’s future profits.
But it’s worth examining the other side of the coin. There are few things people value more than their health. U.S. governmental agencies place the value of a single life at somewhere between $1 million and $10 million. The commonly used quality-adjusted life year metric offers valuations that range from $50,000 to upward of $300,000 per incremental year of life.
It therefore follows that the trivial sums users of genetic-testing kits might derive from a system of data property rights would likely be dwarfed by the value they would enjoy from improved medical treatments. A strong case can be made that policymakers should prioritize advancing the emergence of new treatments, rather than attempting to ensure that consumers share in the profits generated by those potential advances.
These debates drew increased attention last year, when 23andMe signed a strategic agreement with the pharmaceutical company Almirall to license the rights related to an antibody Almirall had developed. Critics pointed out that 23andMe’s customers, whose data had presumably been used to discover the potential treatment, received no monetary benefits from the deal. Journalist Laura Spinney wrote in The Guardian newspaper:
23andMe, for example, asks its customers to waive all claims to a share of the profits arising from such research. But given those profits could be substantial—as evidenced by the interest of big pharma—shouldn’t the company be paying us for our data, rather than charging us to be tested?
In the deal’s wake, some argued that personal health data should be covered by property rights. A cardiologist quoted in Fortune magazine opined: “I strongly believe that everyone should own their medical data—and they have a right to that.” But this strong belief, however widely shared, ignores important lessons that law and economics has to teach about property rights and the role of contractual freedom.
Why Do We Have Property Rights?
Among the many important features of property rights is that they create “excludability,” the ability of economic agents to prevent third parties from using a given item. In the words of law professor Richard Epstein:
[P]roperty is not an individual conception, but is at root a social conception. The social conception is fairly and accurately portrayed, not by what it is I can do with the thing in question, but by who it is that I am entitled to exclude by virtue of my right. Possession becomes exclusive possession against the rest of the world…
Excludability helps to facilitate the trade of goods, offers incentives to create those goods in the first place, and promotes specialization throughout the economy. In short, property rights create a system of exclusion that supports creating and maintaining valuable goods, services, and ideas.
But property rights are not without drawbacks. Physical or intellectual property can lead to a suboptimal allocation of resources, namely market power (though this effect is often outweighed by increased ex ante incentives to create and innovate). Similarly, property rights can give rise to thickets that significantly increase the cost of amassing complementary pieces of property. Often cited are the historic (but contested) examples of tolling on the Rhine River or the airplane patent thicket of the early 20th century. Finally, strong property rights might also lead to holdout behavior, which can be addressed through top-down tools, like eminent domain, or private mechanisms, like contingent contracts.
In short, though property rights—whether they cover physical or information goods—can offer vast benefits, there are cases where they might be counterproductive. This is probably why, throughout history, property laws have evolved to achieve a reasonable balance between incentives to create goods and to ensure their efficient allocation and use.
Personal Health Data: What Are We Trying to Incentivize?
There are at least three critical questions we should ask about proposals to create property rights over personal health data.
What goods or behaviors would these rights incentivize or disincentivize that are currently over- or undersupplied by the market?
Are goods over- or undersupplied because of insufficient excludability?
Could these rights undermine the efficient use of personal health data?
Much of the current debate centers on data obtained from direct-to-consumer genetic-testing kits. In this context, almost by definition, firms only obtain consumers’ genetic data with their consent. In western democracies, the rights to bodily integrity and to privacy generally make it illegal to administer genetic tests against a consumer or patient’s will. This makes genetic information naturally excludable, so consumers already benefit from what is effectively a property right.
When consumers decide to use a genetic-testing kit, the terms set by the testing firm generally stipulate how their personal data will be used. 23andMe has a detailed policy to this effect, as does Family Tree DNA. In the case of 23andMe, consumers can decide whether their personal information can be used for the purpose of scientific research:
You have the choice to participate in 23andMe Research by providing your consent. … 23andMe Research may study a specific group or population, identify potential areas or targets for therapeutics development, conduct or support the development of drugs, diagnostics or devices to diagnose, predict or treat medical or other health conditions, work with public, private and/or nonprofit entities on genetic research initiatives, or otherwise create, commercialize, and apply this new knowledge to improve health care.
Because this transfer of personal information is hardwired into the provision of genetic-testing services, there is space for contractual bargaining over the allocation of this information. The right to use personal health data will go toward the party that values it most, especially if information asymmetries are weeded out by existing regulations or business practices.
Regardless of data property rights, consumers have a choice: they can purchase genetic-testing services and agree to the provider’s data policy, or they can forgo the services. The service provider cannot obtain the data without entering into an agreement with the consumer. While competition between providers will affect parties’ bargaining positions, and thus the price and terms on which these services are provided, data property rights likely will not.
So, why do consumers transfer control over their genetic data? The main reason is that genetic information is inaccessible and worthless without the addition of genetic-testing services. Consumers must pass through the bottleneck of genetic testing for their genetic data to be revealed and transformed into usable information. It therefore makes sense to transfer the information to the service provider, who is in a much stronger position to draw insights from it. From the consumer’s perspective, the data is not even truly “transferred,” as the consumer had no access to it before the genetic-testing service revealed it. The value of this genetic information is then netted out in the price consumers pay for testing kits.
If personal health data were undersupplied by consumers and patients, testing firms could sweeten the deal and offer them more in return for their data. U.S. copyright law covers original compilations of data, while EU law gives 15 years of exclusive protection to the creators of original databases. Legal protections for trade secrets could also play some role. Thus, firms have some incentives to amass valuable health datasets.
But some critics argue that health data is, in fact, oversupplied. Generally, such arguments assert that agents do not account for the negative privacy externalities suffered by third-parties, such as adverse-selection problems in insurance markets. For example, Jay Pil Choi, Doh Shin Jeon, and Byung Cheol Kim argue:
Genetic tests are another example of privacy concerns due to informational externalities. Researchers have found that some subjects’ genetic information can be used to make predictions of others’ genetic disposition among the same racial or ethnic category. … Because of practical concerns about privacy and/or invidious discrimination based on genetic information, the U.S. federal government has prohibited insurance companies and employers from any misuse of information from genetic tests under the Genetic Information Nondiscrimination Act (GINA).
But if these externalities exist (most of the examples cited by scholars are hypothetical), they are likely dwarfed by the tremendous benefits that could flow from the use of personal health data. Put differently, the assertion that “excessive” data collection may create privacy harms should be weighed against the possibility that the same collection may also lead to socially valuable goods and services that produce positive externalities.
In any case, data property rights would do little to limit these potential negative externalities. Consumers and patients are already free to agree to terms that allow or prevent their data from being resold to insurers. It is not clear how data property rights would alter the picture.
Proponents of data property rights often claim they should be associated with some form of collective bargaining. The idea is that consumers might otherwise fail to receive their “fair share” of genetic-testing firms’ revenue. But what critics portray as asymmetric bargaining power might simply be the market signaling that genetic-testing services are in high demand, with room for competitors to enter the market. Shifting rents from genetic-testing services to consumers would undermine this valuable price signal and, ultimately, diminish the quality of the services.
Perhaps more importantly, to the extent that they limit the supply of genetic information—for example, because firms are forced to pay higher prices for data and thus acquire less of it—data property rights might hinder the emergence of new treatments. If genetic data is a key input to develop personalized medicines, adopting policies that, in effect, ration the supply of that data is likely misguided.
Even if policymakers do not directly put their thumb on the scale, data property rights could still harm pharmaceutical innovation. If existing privacy regulations are any guide—notably, thepreviously mentioned GDPR and CCPA, as well as the federal Health Insurance Portability and Accountability Act (HIPAA)—such rights might increase red tape for pharmaceutical innovators. Privacy regulations routinely limit firms’ ability to put collected data to new and previously unforeseen uses. They also limit parties’ contractual freedom when it comes to gathering consumers’ consent.
At the margin, data property rights would make it more costly for firms to amass socially valuable datasets. This would effectively move the personalized medicine space further away from a world of permissionless innovation, thus slowing down medical progress.
In short, there is little reason to believe health-care data is misallocated. Proposals to reallocate rights to such data based on idiosyncratic distributional preferences threaten to stifle innovation in the name of privacy harms that remain mostly hypothetical.
Data Property Rights and COVID-19
The trade-off between users’ privacy and the efficient use of data also has important implications for the fight against COVID-19. Since the beginning of the pandemic, several promising initiatives have been thwarted by privacy regulations and concerns about the use of personal data. This has potentially prevented policymakers, firms, and consumers from putting information to its optimal social use. High-profile issues have included:
Each of these cases may involve genuine privacy risks. But to the extent that they do, those risks must be balanced against the potential benefits to society. If privacy concerns prevent us from deploying contact tracing or green passes at scale, we should question whether the privacy benefits are worth the cost. The same is true for rules that prohibit amassing more data than is strictly necessary, as is required by data-minimization obligations included in regulations such as the GDPR.
If our initial question was instead whether the benefits of a given data-collection scheme outweighed its potential costs to privacy, incentives could be set such that competition between firms would reduce the amount of data collected—at least, where minimized data collection is, indeed, valuable to users. Yet these considerations are almost completely absent in the COVID-19-related privacy debates, as they are in the broader privacy debate. Against this backdrop, the case for personal data property rights is dubious.
Conclusion
The key question is whether policymakers should make it easier or harder for firms and public bodies to amass large sets of personal data. This requires asking whether personal data is currently under- or over-provided, and whether the additional excludability that would be created by data property rights would offset their detrimental effect on innovation.
Swaths of personal data currently lie untapped. With the proper incentive mechanisms in place, this idle data could be mobilized to develop personalized medicines and to fight the COVID-19 outbreak, among many other valuable uses. By making such data more onerous to acquire, property rights in personal data might stifle the assembly of novel datasets that could be used to build innovative products and services.
On the other hand, when dealing with diffuse and complementary data sources, transaction costs become a real issue and the initial allocation of rightscan matter a great deal. In such cases, unlike the genetic-testing kits example, it is not certain that users will be able to bargain with firms, especially where their personal information is exchanged by third parties.
If optimal reallocation is unlikely, should property rights go to the person covered by the data or to the collectors (potentially subject to user opt-outs)? Proponents of data property rights assume the first option is superior. But if the goal is to produce groundbreaking new goods and services, granting rights to data collectors might be a superior solution. Ultimately, this is an empirical question.
As Richard Epstein puts it, the goal is to “minimize the sum of errors that arise from expropriation and undercompensation, where the two are inversely related.” Rather than approach the problem with the preconceived notion that initial rights should go to users, policymakers should ensure that data flows to those economic agents who can best extract information and knowledge from it.
As things stand, there is little to suggest that the trade-offs favor creating data property rights. This is not an argument for requisitioning personal information or preventing parties from transferring data as they see fit, but simply for letting markets function, unfettered by misguided public policies.
RSS - Posts