3.2.3.1 Judicial Review

We begin with “classic” judicial review. Of course, Member States would have had courts and other adjudicators at their disposal. However, administrifying platforms at the Union level just to hand over a crucial procedural mechanism – governing the review procedures when it came to platforms – to heterogenous and differently paced national judiciaries would not have matched the DSA’s overall thrust. The DSA seeks to harmonize, not to pluralize.Footnote ³⁸ Further, some interviewees who had dealt with the legislative procedure at an early stage pointed to a widespread understanding among EU lawmakers that national judicial procedures were of varying quality and, perhaps even more important, too slow.Footnote ³⁹ National judicial processes arguably cannot offer the accessibility and speed the Commission wanted. Of course, anyone who thinks their content has been wrongly removed by a social media platform may initiate civil litigation (and some people do). But many of these judgments, for example decisions by Germany’s Bundesgerichtshof also do not offer much more than the DSA does in this regard.Footnote ⁴⁰ Again, the chosen normative tools are those known from administrative law: nondiscrimination, a duty to give reasons, perhaps a right to be heard, and perhaps proportionality.Footnote ⁴¹ Either way, litigating before national courts is time-consuming and expensive and thus unsuited to achieving what the Commission’s lawmakers wanted.

Further, demanding that EU courts deal with the expected mass of content moderation-related litigation was also a nonstarter. The European Court of Justice (ECJ)’s docket has a huge backlog and adjudicatory protection against EU agencies has already been partially outsourced to so-called Boards of Appeal.Footnote ⁴²

3.2.3.2 Internal Review

The second option is contending decisions directly against the decision-maker. The right to go back to the administrative actor to demand a second look and voice criticism is a classic one in administrative law.Footnote ⁴³ Thus, the Commission included it in the DSA, which offers individuals access to an “internal complaint-handling system.” According to Article 20(4) of the DSA, social media platforms

This provision codifies many ideas swirling around and would warrant its own chapter, which is beyond the scope of this study.Footnote ⁴⁴ The administrifying intent of this provision is clear. It subjects platforms to the principle of non-discrimination and procedural consistency because platforms are prohibited from acting arbitrarily. Although not new,Footnote ⁴⁵ this positive codification might have huge practical effects as it requires platforms to develop a normative corpus to safeguard the consistency of rulings. Given the millions of decisions that content moderation platforms make all the time (and the myriad design choices and rulemaking choices coloring every aspect of user behavior) “nonarbitrariness” requires a system of general and hierarchically structured rules to guide “downstream,” user-facing decisions. Perhaps, this provision will be identified as the nucleus of platform-oriented administrative law in the future. But that is not our concern for now.

However, internal complaint-handling systems have one irredeemable flaw. They are not independent. Instead, the “complaint handler” is internal and keeps the dispute within the dyadic relationship of individual user and platform. The main function of such dyadic, internal complaint-handling mechanisms is to enhance the quality of decisions and to give decision-makers a second chance to undo mistakes before they become expensive, time-consuming, and potentially embarrassing. Such a forum for exercising “voice” is critical for users and, in principle, the easiest and most efficient form of reconciling disputes prior to the option of expanding dispute resolution into a triad. However, the dyadic nature of internal complaint-handling has clear limitations. Most importantly, the decision-maker on behalf of the governance structure – in our case the platform – does not substantially change. Perhaps other bodies further up the ladder decide complaints but those will be part of the same organizational entity. The internal units deciding those complaints are still entrenched in the platforms’ hierarchy, must report to superior executives, and are subject to corporate policies, chief of which is the pursuit of profit.

Article 20 of the DSA seeks to remedy this entrenchment by stipulating that internal complaints must be handled in a nondiscriminatory, diligent, and nonarbitrary fashion. But what these terms mean remains murky. We do not know the criteria for “diligence,” nor does the DSA stipulate the legal yardsticks against which the platforms are “reasonably” supposed to measure their nonarbitrariness. Surely, platform decisions should be effective and not violate EU fundamental rights. But is cost efficiency, for example, a reasonable or an arbitrary consideration when handling internal appeals?

At this point, we may realize that one notion that would have provided a legal basis for better decision-making is missing: impartiality. In the previous chapter I showed that from a functional as well as a historical perspective, impartiality, and not (only) independence, is arguably the characterizing feature of good adjudication.Footnote ⁴⁶ Traditional, public administrative actors are expected to act impartially in their relations with individuals.Footnote ⁴⁷ For example, the ECJ expects the European Commission to “examine carefully and impartially all the relevant aspects of the individual case, the right of the person concerned to make his views known and to have an adequately reasoned decision.”Footnote ⁴⁸ Platforms’ internal complaint-handling systems do not have to be impartial in that public-law sense. Article 20 of the DSA makes no reference to impartiality. It does reference nonarbitrariness though. However, the notion of “nonarbitrariness” can only be interpreted in the economic context in which platforms operate and to which the DSA should apply. Although the EU legislators are trying to an extent to administrify platforms, the latter will still be private, profit-oriented enterprises. Profit orientation, a preference for quick fixes,Footnote ⁴⁹ and economies of scale saturate platforms’ internal structures and the habits and attitudes of most social media platforms and employees.Footnote ⁵⁰ For a profit-oriented company, handling internal complaints as cheaply as possible is not an “arbitrary” decision at all. In terms of platforms’ legally protected freedom to do business, handling internal complaints cheaply – even if that lowers the quality of their services – might be a rational and even legally protected choice. Thus, platforms are expected to consider aspects like cost efficiency when internally handling appeals, if not by law, then in practice. We know from testimony by Frances Haugen, a former executive-turned whistleblower, that Meta executives (and arguably those of other platforms) regularly decide against improving administrative procedures if they consider these too costly to implement:

In general, the regulation also seeks to accommodate the business interests of platforms – a balancing act that seems to be constitutionally demanded by the provisions of Article 16 of the Charter of Fundamental Rights (CFR).Footnote ⁵² For example, the DSA should “provide … the conditions for innovative digital services to emerge and to scale up in the internal market.”Footnote ⁵³ In practice, platforms already allow for internal appeals – with mixed results. To sum up, internal complaint-handling is a crucial brick in the edifice of administrative justice, hence its appearance in the DSA. Against this backdrop, the DSA’s approach to granting considerable leeway to platforms regarding how to implement and interpret procedural and other normative targets like “nonarbitrary internal complaint-handling” appears a risky one. We will further reflect on this in the chapter dealing with outcomes.Footnote ⁵⁴

To conclude, the power imbalance between platforms and users remain unchanged in internal complaint-handling. Therefore, contending platform decisions solely in front of internal complaint-handlers offers only limited accountability. So, introducing internal complaint-handling systems for platforms may offer some – but arguably not enough – legitimacy to the EU’s desire to be a regulator with teeth.

3.2.3.3 ODSs as Middle Ground

The shortcomings of the two previous options – real courts are complicated and having no adjudication at all risks underwhelming stakeholders – have given rise to a kind of institutional hermaphrodite. The Commission proposed establishing independent, adjudicatory bodies – the ODSs – which some commentators at the time have quite reasonably interpreted as “de facto courts.”Footnote ⁵⁵ Out-of-court dispute settlement bodies are independent bodies set up to decide on disputes between individuals and platforms. Out-of-court dispute settlement bodies are private, yet publicly mandated. They base their decisions either on platforms’ terms of service or on the acquis of applicable EU and Member State law.Footnote ⁵⁶ Platforms are legally obliged to participate in proceedings before ODSs. Initially, the bodies’ decisions were designed to be binding, which would have made ODSs specialized adjudicators controlling content moderation. But as we will see later, the Council scrapped ODSs’ adjudicative authority. In the eventually passed version of the legislation, their decisions are not binding (although, the EU seems to be pinning its hopes on other enforcement mechanisms).

Politically, ODSs offered – at least as a result of the authority initially invested in them – legitimacy benefits to the EU because they would have allowed the EU to pose as a guarantor of individual rights online. By establishing ODSs, the policymakers’ goal was that EU citizens could “easily and effectively contest certain decisions of online platforms that negatively affect them.”Footnote ⁵⁷ This is indeed a global legislative novelty.Footnote ⁵⁸ As argued in the previous chapter, individual rights adjudication is a revered sociocultural reference point and an institutionalized element of legitimate administrative power structures. Infusing such administrative-law principles into private power structures will naturally be accompanied by some sort of individual rights adjudication. Correspondingly, legal experts welcomed the Commission’s initial proposal “for aiming to introduce a system of out-of-court dispute resolution that promises the expedient settlement of conflicts involving harmful user-generated content.”Footnote ⁵⁹

Remarkably, ODSs are not the first adjudicatory institutions the EU has set up to address power imbalances in private governance. Particularly when it came to consumer protection, Directive 2013/11/EU established a system of equally nonbinding ADR bodies.Footnote ⁶⁰ As we will discuss in Section 3.3.3, the Commission proposal for ODSs partly emulates the ADR Directive.Footnote ⁶¹ But one legal scholar, who had welcomed the Commission proposal for ODSs in 2020, criticized the ADR Directive. In a 2014 article, Gerhard Wagner argued “ADR is an inadequate means of law enforcement” and “European consumer law is by no means an area so simple as to render adjudication unimportant. … [I]t is all the more surprising that the EU lawmakers put so much faith in alternative dispute resolution. How would a non-court entity be able to apply the law more easily than a court would?”Footnote ⁶²

However, that the Commission would draw from earlier experiences with ADR when devising the DSA can also be explained through contextual factors. Several of the Commission agents who drafted Articles 20 and 21 of the DSA had previously worked in the context of the consumer ADR Directive.Footnote ⁶³ Other role models that might have influenced the Commission’s thinking were arguably Article 17(9) subparagraph 2 sentence 2 Digital Single Market Directive, which mandated Member States to “ensure that out-of-court redress mechanisms are available for the settlement of disputes,” Article 12 of the P2B Directive, which nudges online intermediaries toward participating in mediation procedures, and §3(c) of the German NetzDG, which also briefly established voluntary mediation services for (some) disputes between platforms and users in Germany.Footnote ⁶⁴ Yet, none of these earlier provisions foresaw binding adjudication and mandatory participation by platforms. How the EU wove these private- and public-law perspectives together will be detailed below.Footnote ⁶⁵ Before diving into the details of ODSs’ origin, we should first look at the DSA’s overall characteristics.

3.2.4.1 Procedure

Many of the procedures the DSA concentrates on are individualized in that they focus on the relationship between person and platform – and less on platforms’ structural governance mechanisms that inform individualized decisions.Footnote ⁷³ Seen through the lens of rulemaking and rule enforcement, the DSA proceduralizes rule enforcement but not rulemaking. For example, platforms must establish internal complaint-handling mechanisms (Article 20) and are subject to the rulings of the relevant out-of-court dispute settlement bodies (Article 21). Here, the DSA seeks to empower individuals to use individual rights as power mediators vis-à-vis the platforms on two levels.

Beyond the individual level, the DSA introduced several, more general and structural, procedures to enhance the transparency of platforms. For example, extensive platforms “shall diligently identify, analyze and assess” whether they pose “systemic risks” for aspects such as “foreseeable negative effects on civic discourse and electoral process, and public security.”Footnote ⁷⁴ Further, platforms are subject to auditing (Article 37) and transparency obligations regarding their use of data (Articles 39, 40, and 42) and must establish a “compliance function” (Article 41). In contrast, the DSA offers very few specific, standard-related criteria in its definitions of what practices or normative judgments are allowed or prohibited. For example, Articles 34 and 35 leave standard-setting for matters like auditing (Article 34(1)(d)) to “voluntary industry standards” and equally “self”-devised codes of conduct.

3.2.4.2 Design

However, the DSA’s reformist agenda arguably misses out on two crucial aspects: rulemaking and inner-platform decision-making prior to enforcement (in other words, the design of platforms). The previous chapter briefly sketched out how content moderation relies not only on the binary enforcement of company-made rules but also on the design, programming, and infrastructure employed by platforms vis-à-vis their users.Footnote ⁷⁵ Initially, aspects of platform design or internal decision-making, although referenced,Footnote ⁷⁶ were not a prominent feature in the Commission proposal.Footnote ⁷⁷ Later, perhaps reflecting evolving academic views on content moderation, a newly introduced Article 25(1) of the DSA sketched a broad design framework for platform–user relationships prior to questions of formal enforcement. This particular clause stipulated that social media platforms were to “design, organize or operate their online interfaces” in no way that “deceives or manipulates” or “materially distorts or impairs the ability” of individual users “to make free and informed decisions.” According to Article 25(3) of the DSA, the Commission may publish further guidelines specifying design principles. For the moment, the legal status of those design “principles” remains vague. What counts as deception and impairment of “free” decision-making is still unclear. After all, building a doctrinally sound legal framework that protects the exercise of free will itself would be a Sisyphean task.Footnote ⁷⁸ Nonetheless, the Commission could in theory fine platforms for violating these design principles.Footnote ⁷⁹ A cynic might even say that any social media platform’s design is anyway geared toward deceiving, manipulating, and impairing the exercise of free will on the part of its users.Footnote ⁸⁰ Algorithms lure users into spending more time on-screen. Based on mined behavioral data the companies anticipate, encourage or even produce needs among their customers. Further, social media usage can lead to an increase in users experiencing a host of negative effects, ranging from multiplying cases of depression and anorexia to growing incidents of suicide.Footnote ⁸¹ From a rule of law perspective, such overly broad and in practice unrealizable design criteria perhaps should not be “enforceable” in the event of their absence by fining the companies up to 6 percent of their combined global annual turnover.Footnote ⁸²

3.2.4.3 Rulemaking

The second crucial aspect the DSA does not regulate is rulemaking. Platform rulemaking – that is, platforms’ unilateral generating of norms that delineate what kinds of communication they allow – is a key aspect of content moderation.Footnote ⁸³ The Commission has pointed out: “platforms set the rules of the game, without effectively mitigating the risks and the societal and economic harm they cause.”Footnote ⁸⁴ And yet the DSA does not seek to weigh in on platform rulemaking. Specifically, it does not limit platforms’ power to unilaterally make and on their own behalf change these “rules of the game.”Footnote ⁸⁵ Article 14(4) of the DSA obliges platforms to “act in a diligent, objective and proportionate manner in applying and enforcing the … [terms of service], with due regard to the rights and legitimate interests of all parties involved, including the applicable fundamental rights.” Martin Eifert and others argue that this means the DSA grants platforms “unfettered discretion” to exercise private regulation of public discourse.Footnote ⁸⁶

The DSA’s due diligence obligations only concern, as already mentioned, individual application and enforcement – not rulemaking. From a public-law perspective, binding these strands of individual enforcement decisions to individual rights seems to make intuitive sense. If decision-makers must pay “due regard” to fundamental rights when enforcing terms of service, the problem should disappear.Footnote ⁸⁷

In an ideal world, Article 14(4) of the DSA requires platforms to engage in a long-term balancing act, which should – the theory goes – prevent any infringements of individual rights. However, the reality is that balancing fundamental rights is complex, and content moderation decisions are made within constraints of time and resources. To cite just two numbers: between January and March 2024, YouTube deleted almost 8 million videos and more than 1.4 billion comments worldwide.Footnote ⁸⁸ Human moderators, usually low-paid and based in South-East Asia, normally have only a few seconds per enforcement decision.Footnote ⁸⁹ This may also work similarly in the case of internally handled appeals, which might be dealt with by different teams that would, perhaps, tend to be working in slightly better material and institutional conditions. In short, the reality of content moderation appears irreconcilable with the ideal premised in Article 14(4) of the DSA.

Also, in its introductory “recitals,” the DSA text does not explain how platforms are supposed to pay “due regard” to fundamental rights when restricting communication.Footnote ⁹⁰ The idea of people being required to consciously balance conflicting fundamental rights positions for every moderation-related decision is an illusion. In turn, the DSA fails to define procedural obligations – except that terms of service must be published and summarized in the relevant languages – that would inform decision-making and rulemaking processes.

Given our emulation-oriented framework, this key provision will probably only lead to the lip-service of ceremonial compliance. On paper, establishing a kind of ceremonial commitment to “duly regard” fundamental rights in all these decisions will be easy. There will be briefings, commitments, training courses, and reports aiming to enhance moderators’ “awareness” of fundamental rights. However, none of these extras will enable moderators to take more time over individual enforcement decisions, which ought to be the baseline requirement for any genuine attempts to balance the fundamental rights of different protagonists – including potentially several users and, we should not forget, the platform itself – against each other. To put it bluntly, Article 14 of the DSA may serve as a sedative for the widespread debate about whether social media platforms ought to be “bound” in any sense by fundamental rights.Footnote ⁹¹ But it most likely lacks precision and force to effectuate meaningful change.

What is truly astonishing is the following: The Commission not only seemingly knew but even openly acknowledged these limitations from the very beginning of our whole saga in winter 2020. Concretely, the Commission assumes that implementing Article 14(4) of the DSA will require so little effort from platforms that it will come at virtually no cost. This admission can be found buried in an impact assessment published in 2020 alongside the initial Commission proposal. As argued above, the effective implementation of Article 14(4) and other key provisions would necessitate substantial organizational, normative, and procedural overhauls within platforms. By definition, such overhauls would be costly, especially considering the high-minded balancing objectives embedded in Article 14(4). After all, driving significant governance reforms at platforms is the core purpose of the DSA. Yet, in that impact assessment released alongside the initial proposal, the Commission concedes that for most companies the costs for engaging in public-minded moderation practices would,

To reiterate, the Commission asserts that implementing key provisions of the DSA will be no more expensive than “current operations.” But if that were true, why have platforms not already adopted the practices the DSA now mandates? If exemplary moderation costs the same as mediocre moderation, why has the state of content moderation in Europe become such a pressing issue that the European Union felt compelled to pass the DSA? Conversely, if the DSA’s supposedly “groundbreaking” demands require so little change at the platforms, what do we need the DSA for? The best things in life are free, but good content moderation certainly is not.

Automated moderation, on the other hand, does not “consciously” weigh anything but builds on specific, preexisting, computed rules and probabilistic patterns generated from large data sets.Footnote ⁹³ Although hailed as a solution to content moderation’s problem of scale, automated decision-making seems not (yet) equipped to “understand” the apparent “meanings” of human communication, let alone engage in a balancing of fundamental rights.Footnote ⁹⁴ Even though recent advances in large language models are indeed impressive, at the time of this writing, neither the quality of the applied normative material nor the employed algorithms seem to have yet salvaged the described problems. Anyone who studied administrative actors knows how difficult maintaining decisional quality can be even with human operators. If companies eventually manage to out-engineer this problem – and that might be possible, to an extent – maintaining intrinsically anthropomorphic ideals like “balancing” and “proportionality” in a system that runs through code rather than humans will be ever more challenging.Footnote ⁹⁵

In sum, content moderation “decisions” are not so much individualized, case-by-case decisions but repeated applications of specific rules and other “upstream” normative decisions.Footnote ⁹⁶ As recent research corroborates, the crucial normative questions of content moderation are not hidden in individual rule applications but entrenched in the vast landscape of specific rules, guidelines, design choices, and other normative material made by platforms.Footnote ⁹⁷

It appears questionable whether individual rights are the (only) appropriate tool to tackle this problem. If individual decisions are just reflections of previously generated rules and other normative choices embedded in a platform’s hardware, directing appeals against individual enforcement decisions is pointless. Consequently, Evelyn Douek even speaks of “accountability theater” in that regard.Footnote ⁹⁸

3.3.1.1 Informal Consultations

Expert theorization, often conveyed in informal consultations with policymakers, is a driving force behind emulation.Footnote ¹¹¹ Emulation refers to a way by which preexisting role models are reproduced in policymaking and organizational design.Footnote ¹¹² Policymakers are usually acting within time and resource constraints.Footnote ¹¹³ They therefore strive to include experts’ insights and proposals in policymaking. As a result, experts will theorize based on preexisting assumptions and often advise policymakers to adopt policy choices or organizational designs that work in other contexts. In other words, experts will often advise their clients to emulate.

The role of expert advice is particularly salient for the European Commission. Tasked to formulate regulatory proposals in vastly complex fields, the Commission engages in extensive expert consultations before drafting a proposal.Footnote ¹¹⁴ Although the Commission also conducts open and public consultations, internal and relatively nontransparent “behind the scenes” discussions tend to precede public consultations. As empirical research has evidenced, internal, circumscribed, and “capacity-building” consultations with experts and other stakeholders outnumber public consultations, especially when it comes to high-profile policy proposals like the DSA.Footnote ¹¹⁵ Typically, Commission consultations may include “expert groups, as well as seminars, workshops, or conferences which are institutionalized to a minimum degree,”Footnote ¹¹⁶ with only a few of these accessible to the public. For the DSA, the Commission held public consultations between June and September 2020, just months before publishing the final proposal in December.Footnote ¹¹⁷ It received more than 2,800 pieces of feedback in these open consultations.Footnote ¹¹⁸ Arguably, none of these eleventh-hour consultations had any meaningful effect on the proposal. Big questions are usually cleared way before such a proposal enters the public stage.

In contrast, informal and internal discussions precede public consultations because the former lay the groundwork for the ideas and texts subsequently discussed in the latter.Footnote ¹¹⁹ This was also the case for the DSA. In supplementary material to its proposal, the Commission said it had “gathered expertise and views through targeted consultations and engagement activities, including a series of workshops, conferences, interviews with experts and judges, consultations of the Expert Group on e-commerce, as well as numerous bilateral meetings and analysis of ad hoc position and research papers from organizations, industry representatives, civil society and academia.”Footnote ¹²⁰

Such informal discussions are typically only open to invited experts, often working in a particular policy field.Footnote ¹²¹ In practice, such early, informal influence is crucial and perhaps even decisive. As Päivi Leino-Sandberg notes, based on interviews with members of the Commission’s Legal Service: “The real legal battles within the Commission take place prior to the adoption of a proposal.”Footnote ¹²² Within the Commission, drafting a proposal relies on extensive “internal legal consultations [that] involve exchanges between officials at the level of technical experts.”Footnote ¹²³ During these discussions, an intricate dance of expertise, personality, political agenda, the relevant Directorate General’s ideas, and the Commission’s Legal Service law-related (and sometimes political) observations coalesce into the details of the Commission proposal.Footnote ¹²⁴

3.3.1.2 ‘Compulsory and Compensated ADR’

For ODSs, we can trace the influence of expert theorizations even beyond the discourse-oriented influence described above. One specific expert theorization was apparently a strong influence on the inclusion of and particular design of ODSs in the Commission proposal. That expert theorization can be traced to Martin Husovec, a law professor. Professor Husovec amicably agreed to be interviewed for this book. In the mid-2010s, when the Commission was contemplating reform of its eCommerce Directive, Husovec pitched what he called “compulsory and compensated ADR” to Commission staff.Footnote ¹²⁵ That was “well before” the DSA proposal was planned out in detail.Footnote ¹²⁶ Husovec stressed that he “would not want to paint the picture” that he had come up with the initial idea for out-of-court dispute settlement bodies. And yet he was apparently the first expert to pitch to the Commission the notion of compulsory and compensated ADR bodies as potential controllers of platform power.Footnote ¹²⁷ In the early 2010s, the number of lawmakers, academics, and lobbyists working on platform regulation was not that large.Footnote ¹²⁸ For example, conference programs from the time show the relevant people from the Commission and academics sharing panels, indicating that some contacts dated back years.Footnote ¹²⁹ At the same time, the Commission was (internally) looking at ADR as a potential role model for overhauling the eCommerce Directive.Footnote ¹³⁰ As Husovec had worked extensively on intermediary liability, he and a colleague designed an experimental study to empirically test the governance-improving effects of what would later become ODSs. In the final publication of this study (but also in the earlier published preprint version) Husovec and Lenka Fiala, his coauthor, wrote:

Simply put, ODSs seem to be emulation in its purest form. The EU needed a publicly legitimate but also realistic solution for a crucial policy problem. With the help of advice from experts it used models drawn from other contexts – here, consumer protection and, in the case of Husovec and Fiala, copyright dispute resolution – and applied them to a new context. Eventually, the emulation translates into both a policy and a formal structure for enacting that policy.

To operationalize these theorizations, the Commission consulted with Husovec and other experts.Footnote ¹³² For example, buried in supplementary legislative materials, we find that Husovec organized a “small-scale workshop [for Commission agents] about the future of liability of digital platforms in Europe.”Footnote ¹³³ Apparently, this workshop focused on the drafting of Articles 17 and 18 of the Commission proposal, which dealt with internal complaint-handling mechanisms and, crucially, ODSs and would later become Articles 20 and 21 DSA. According to the Commission, “[t]he event’s goal was to share the latest academic research with the EU officials and discuss potential solutions for the reform of the [eCommerce Directive], including drafting suggestions for the provisions related to intermediary liability and notice-and-action mechanisms.”Footnote ¹³⁴

This is remarkable. We must assume that not only the broad idea but the very formulation of the text that would later become Article 21 was discussed with outside experts in these informal consultations. “Compulsory and compensated ADR,” as proposed by Husovec and Fiala, should incentivize delegated enforcers like platforms to take a more nuanced approach to blocking content, especially with automated tools. It builds on the standard picture of content moderation as comprising a continual string of individual decisions.Footnote ¹³⁵ Husovec argued that compulsory and compensated ADR would help align private companies’ decision-making activities with fundamental rights.Footnote ¹³⁶ Interestingly, the professor’s background was not in public law but private law, with a focus on intermediary liability and law and economics.Footnote ¹³⁷

Compulsory and compensated ADR was proposed by Husovec and Fiala to work as follows. If a platform decides against a user, for example, by deleting content, that person can go to an ADR body to seek redress. Platforms are obliged to take part in ADR proceedings, that is to say, ADR becomes compulsory for platforms. If they lose, platforms must pay a fine, that is to say, the proceedings are compensated.Footnote ¹³⁸ Thus, the key function of “compulsory and compensated” ADR proceedings is to inflict costs upon misgoverning online platforms. These costs would then incentivize platforms to improve their governance. In the words of Husovec and Fiala:

The authors argued that inflicting costs is such an essential measure that the adjudicatory body’s design does not matter “as long as the financing structure remains the same.”Footnote ¹⁴⁰ In their model, individual rights adjudication merely serves as a procedural legitimizer of the punitive costs inflicted on the platform in each case. Adjudication determines whether platforms must “comply” and “pay additional fees.”Footnote ¹⁴¹ These additional fees were the crucial, punitive aspect of the model. If enough cases pile up, platforms would in theory be motivated to modify their rulemaking and rule enforcement, as they would presumably prefer to avoid repeatedly paying “additional fees” when the same or similar issues arose in future.Footnote ¹⁴² For profit-oriented actors, cost is an incentive to change their organizational behavior. This aspect is decisive. Coming from a law and economics background, the expert was skeptical as to whether profit-driven “private bureaucracies” would abide by public law-inspired normative targets. When asked about this, Husovec said, “I’m a very practical-minded person, and I just don’t believe that you can have courts to resolve things. I don’t think every company responds to that. They respond to incentives, mostly financial incentives.”Footnote ¹⁴³ According at least to this expert’s theorization, tying individual rights adjudication to costs for platforms would reconcile platforms’ main motivator (profit) with public interests (individual rights-based appeals).

3.3.4.1 Resistance by Member States

In principle, Article 18 of the Commission proposal was said to have a “rather low profile” in interinstitutional negotiations,Footnote ¹⁶⁷ the reason being that this article was viewed by some as “user-centric,” which was accepted as politically expedient by the Commission, the Parliament, and the Council.Footnote ¹⁶⁸ The Council represents the interests of Member States in the EU lawmaking process. Unlike federal bodies such as the US Senate, however, its members are not elected to their positions but are instead sent by their respective national governments. Although, in tripartite meetings of representatives of the EU’s three main arms (known as “trilogues”), there was a clear rift between, on the one side, the Commission and the Parliament and, on the other side, the Council. All the parties involved seemed to agree with the principle of introducing some kind of independent adjudicator.Footnote ¹⁶⁹ Prior to or during negotiations between institutions, experts began to scrutinize and comment on the Commission proposal. Some produced critiques of the notion of an ODS,Footnote ¹⁷⁰ while others welcomed the DSA’s attempt to establish an independent, additional remedy.Footnote ¹⁷¹ However, with the inevitable talk among industry representatives, activists, and Member State officials, it became clear that various (more or less concerted) efforts were at work to undermine ODSs and especially their legal authority.Footnote ¹⁷² Powerful industry actors apparently adopted a stealth strategy that would not expose them to the public eye while they monitored developments and sought back-channels with stakeholders in and outside Member States and EU institutions.Footnote ¹⁷³ In Council-level discussions, ODSs’ jurisdiction broadened in scope. Further, their “jurisdiction” expanded to include appeals by users who were drawing attention to illegal or harmful content.Footnote ¹⁷⁴ The Commission proposal, unlike the Council’s deliberation, had not allowed for such appeals. It was criticized for “turning a blind eye to the victims” of defamation or other violations when appellants were unable for whatever reason to contest platforms’ decisions.Footnote ¹⁷⁵

Yet one major change regarding Article 18 of the Commission proposal did come about as a result of negotiations with the Council. Several Member States, especially some Nordic countries, resisted the Commission’s plan that decisions made by ODSs should have the authority to bind platforms.Footnote ¹⁷⁶ Commission staffers recalled that several Member States, especially in the north and around the Baltic, were raising even constitutional concerns if ODSs were to have the authority to issue legally binding decisions.Footnote ¹⁷⁷ Thus, opposition to ODSs’ ability to make binding decisions was “relatively straightforward” in the Council.Footnote ¹⁷⁸ Hence, at the Council level, the phrase that online platforms “shall be bound by the decision taken by the body” was cut from Article 18(1) of the DSA proposal in the Council’s “General Approach” of November 2021.Footnote ¹⁷⁹ This was a big deal. What once were imagined as “sub-courts” lost the authority to issue binding decisions.

3.3.4.2 Industry Resistance

However, the Member States just mentioned were not the only actors critical of the Commission proposal regarding an ODS’s authority to bindingly adjudicate. Industry groups representing online platforms lobbied against binding decision-making authority for out-of-court dispute settlement bodies. Some of these groups voiced criticism of the potentially binding force of Article 18 of the Commission proposal. For example, Digital Europe, the powerful alliance of large online platforms in Europe, argued in a position paper:

Intriguingly, in conversations with Commission staffers, such examples of industry criticism – or even just conversations with industry representatives – were never mentioned. But even though Article 18 of the Commission proposal was described as “low profile” in the trilogues, lawmakers were presumably aware of the broad industry consensus against potent ODSs. Between December 2019 and October 2020, the Commission held at least 158 informal consultations with industry groups and digital companies to discuss the DSA/DMA package.Footnote ¹⁸¹ Digital Europe, for instance, sent representatives to meet formally with the Commission on three occasions during that period.Footnote ¹⁸²

Alongside the Commission, industry lobbying regarding the DSA/DMA package focused especially on Council representatives.Footnote ¹⁸³ While “informal consultations” between the Council and industry groups are relatively rare (or under-reported), industry groups are in constant exchange with Member State representatives, who then set out and negotiate the Council’s positions. How the Member States’ representatives formulate their positions and the extent to which industry groups approach them (in Brussels or in their respective capitals) is largely hidden from the public. In other words, lobbying of Member States’ agents can sway discussions in the Council and thereby influence EU lawmaking. Yet Member States share little (or no) information regarding whether or how intensively they consult with industry groups or companies. Further, based on numerous background talks with industry representatives and civil society activists, I can assure that several large technology companies lobbied also on their own terms, that is apart from their backing for Digital Europe above, intensely against Article 18 of the DSA proposal coming into law. For example, a particularly interesting alliance formed between one large digital corporation and child rights activists, which gave the former cover not having to be the one criticizing ODSs in public.

Further, Digital Europe representatives were present in at least some of the eleven meetings disclosed between industry, Council, or Member State representatives at the Council level.Footnote ¹⁸⁴ How effective the platforms’ lobbying actually was remains a matter of speculation. However, it seems plausible to assume that there was, perhaps, some effective influencing at work.

3.7.4.1 ODS’s Incentive Structure Is Insufficient

Let us begin with the first issue. Are the costs platforms must cover high enough to incentivize them to change anything in their moderation practices? The idea was to make ODSs compulsory for platforms and to make losing ADR proceedings costly for platforms. Article 21 of the DSA delivers on the first part of this idea. Also, when platforms lose these procedures, they must “bear all the fees charged by the body and shall reimburse the recipient … for any other reasonable expenses that they have paid,” as Article 21(5) of the DSA sets out. Therefore, the “fees” and “any other reasonable expenses” charged by an ODS might incentivize platforms to improve their governance.

However, what the experts initially wanted – and claimed to have experimental evidence for its efficacy – are “relatively high fees” that would be “additional” to the costs of the parties.Footnote ²⁴³ In other words, not only participating in the proceedings but losing the proceedings was supposed to be costly. Essentially, these extra fees in case of losing the proceedings were supposed to function as “fines” and “penalties.”Footnote ²⁴⁴ The expert theorization, as presumably pitched to the Commission, requires more than cost-covering “fees.” However, in the passed version the bulk of the fees occur always, irrespectively of whether the platform loses or not. Such a model, already on a theoretical level, arguably fails to incentivize platforms not to lose proceedings but only to avoid proceedings altogether.Footnote ²⁴⁵ One particularly concerning prospect to avoid ODS proceedings would be to curb moderating in general. Less moderation, fewer proceedings.

Husovec, who participated in a workshop for the Commission on the drafting of what would later become Article 21 DSA and who had pitched “compulsory and compensated ADR” to the Commission therefore distanced himself from the Commission’s restrictive wording that limited fees to “reasonable expenses.”Footnote ²⁴⁶ In contrast, Husovec in an interview for this book stressed that the reasonable-expenses formulation in Article 21 of the DSA was “nothing” he “came up with.”Footnote ²⁴⁷ Further, he argued that one can read Article 21 of the DSA’s reasonable-fees phrase as in fact aligning with his proposal, because it refers to “this extra fine that comes on top of the nominal fee.”Footnote ²⁴⁸ Nevertheless, one could also “read it as just the standard thing like you win the case, so there is some reasonable cost.”Footnote ²⁴⁹ Thus, it seems doubtful whether the current model of mandatory fees will incentivize platforms to change anything in their content moderation.

3.7.4.2 Bereft of Rulemaking, Adjudication Cannot Deliver Accountability

That being said, one may question how viable some of the assumptions implicitly or explicitly made by the Commission and their consulting experts indeed are. It seems, from my perspective, questionable whether individualized incentive models like that proposed by Husovec and Fiala can – even on a theoretical level – incentivize platforms to do the right thing, that is to say, positively reform their content moderation machineries.

Husovec and Fiala claim to “provide … empirical evidence for the potential positive effects of the proposed policy and identifies factors to consider when operationalizing it in practice.”Footnote ²⁵⁰ However, the real-world validity of some of that alleged evidence appears doubtful. That is for one simple reason: platforms have (much) more choices than Husovec and Fiala suggest.Footnote ²⁵¹ In their paper, they argue that platforms have basically two options when losing in front of an ODS: positively improving their content moderation or keeping it as it is, thereby risking more fines in the future. As platforms want to avoid fines, they would choose improving their content moderation, which then enables them to win future procedures. With such a narrowed bouquet of choices, the paper’s encouraging results are hardly surprising.

If only they were realistic. In reality, platforms can resort to myriad other options to avoid losing more cases. They can, for example, blur their normative material so that it is unclear what is allowed and what is not. Where there is legal uncertainty, one is less likely to be held accountable. And indeed, platforms constantly modify their rules, not always adding to their clarity. In simple terms, costly adjudication based on individual cases but with virtually no influence on rulemaking, may incentivize platforms to ex post “legalize” what they are doing, blur the rules, scrap them, or simply curb moderating altogether to avoid cases.

Therefore, in conclusion, we see how adjudication on its own cannot solve these problems. Irrespective of whether a decision is binding or economically incentivizing – unless the DSA reins into the rulemaking and infrastructural set up of platforms, which it does not – most structural reform of content moderation remains entirely at the whim of platforms.

3.8.1.1 ADROIT

In summer 2024, in scorching Malta, the Maltese Digital Services Coordinator certified the European Union’s very first ODS. That ODS is run by a company called RGOAL Ltd., trading as ADROIT – short for alternative dispute resolution of information technologies – with a registered address and physical office located in the building of a five star resort and conference hotel just outside Valletta. ADROIT’s website comes in six languages, including English, French, German, and Italian.

ADROIT can be seen as a paradigmatic example of the traditional ADR-focused vision for ODSs. It mirrors, to an extent, earlier private law inspired models for ODSs that were discussed in the legislative process.Footnote ²⁵³ Founded by a group of Malta-based German lawyers and a Maltese technical auditor with prior experience in legal tech, ADROIT is part of a broader trend of companies addressing compliance issues for online marketplaces, particularly in the realm of online gambling dispute resolution.

What sets ADROIT apart, however, is its emphasis on one crucial element for the future development of ODSs: entrepreneurship. After all, ODSs are not just legal tools. They are businesses in their own right. Under Article 21 of the DSA, platforms are required to participate in ODS proceedings and cover the associated fees. Given the vast user bases of online platforms in Europe, this creates a potentially lucrative market. ADROIT’s model suggests that we are likely to see highly specialized ODSs emerge, focusing on niche areas such as disputes involving demonetized influencers, app store missteps, or the increasingly profitable world of in-app and in-game purchases (with some games themselves functioning as platforms). Furthermore, we might witness the rise of specialized ODSs catering to particular platforms, types of disputes (e.g., defamation, hate speech, cyberbullying), or even political ideologies (conservative, progressive, etc.). The key for legal entrepreneurs is identifying user groups with sufficient grievances to generate a steady stream of complaints – and thereby revenue.

It is important to note that ADROIT, with its roots in traditional consumer ADR, is not focused on complex content moderation issues that involve fundamental rights.Footnote ²⁵⁴ Instead, ADROIT’s core focus lies in resolving transactional disputes – particularly those related to monetary or contractual issues between individuals and platforms.Footnote ²⁵⁵ According to interviews with ADROIT staff, these often involve matters such as disputes over the pricing and functionality of in-app purchases.Footnote ²⁵⁶

Consequentially, ADROIT positions itself as a mechanism between individuals and platforms to “minimize compliance costs” for the latter.Footnote ²⁵⁷ To achieve this, ADROIT retains the right to “agree on tailored pricing packages depending on the size, complexity, and (expected) caseload of individual Online Platforms.”Footnote ²⁵⁸ Moreover, the company intends to enter into “Mutual Administrative Assistance Agreements … with … Online Platforms,” which also cover “case decision fees.”Footnote ²⁵⁹ The company notes that when “proceedings become somewhat repetitive due to a high number of similar cases concerning the same respondent, this may result in lower average processing costs per case.”Footnote ²⁶⁰ The founder of ADROIT, a lawyer, argues that

From multiple angles, this approach is eminently reasonable. Most importantly, ODS decisions lack legal binding force. Thus, persuading and enticing platforms into compliance, rather than coercing them (with what tools, after all?), emerges as a pragmatic strategy. This holds especially true when one adopts a holistic view of what it means to “implement” a decision within sprawling organizational structures. Only when the relevant actors in key positions of power within a platform’s hierarchy grasp and are genuinely persuaded by external feedback will they be inclined to recalibrate internal practices in response. Such recalibration is likely to occur only if platforms perceive compliance as aligning with their own interests. In this light, compliance fundamentally depends on corporations seeing ODSs not as external impositions but as effective problem-solvers for the platforms themselves.

However, this perspective also highlights the diverse ways in which different ODSs interpret their mandates. Some ODSs position themselves as protectors of fundamental rights – perhaps even adversarial advocates for users against platforms. Others, shaped by their institutional backgrounds, take a markedly different approach. In this light, one might interpret the “Mutual Administrative Assistance Agreement” as ADROIT reserving the right to make structural noncompliance with applicable norms less costly for platforms. Given the nonbinding nature of ODS decisions, this strategy appears logical. Yet, making noncompliance cheaper seems, to me, contrary to some of the original intentions of Article 21. This does not necessarily mean such developments would prove detrimental. However, the prospect reveals two critical points. First, it remains an open question whether the carrot is genuinely more effective than the stick when it comes to motivating platforms. Second, if the “stick” was meant to take the form of financial disincentives, further lowering the cost of noncompliance risks extinguishing any lingering hope that ODSs could, through cumulative costs, nudge platforms toward better practices. Remarkably, making procedures cheaper along the way might not even end up being a strategic choice for ODSs but a legal requirement. Article 21(5)(3) demands that the fees charged to the platform “shall in any event not exceed the costs incurred by the body.” If, economics of scale indeed drive down the costs of ODSs’, one could argue that ODSs are even obliged to lower their fees vis-à-vis platforms.

3.8.1.2 User Rights

While the economic realities and motivations discussed are broadly similar among ODSs, they diverge notably in their orientation toward public law. In this regard, User Rights, the first ODS certified in Germany, represents a distinct model. Certified a few months after ADROIT, User Rights stems from a fundamentally different context, cofounded by a former Oversight Board staffer – an intriguing point of intersection between the two case studies. According to its website and internal documents, including initial decisions I reviewed, User Rights conceptualizes its role as almost entirely driven by public law principles. It ambitiously describes its mission as

User Rights further emphasizes that platforms are obliged to “cooperate in good faith” with its decisions. The decisions I reviewed – particularly one in a more “complex” case – read much like court rulings, notably resembling judgments by a German Verwaltungsgericht, or administrative court (unsurprisingly so as User Right’s founders are German-educated lawyers).Footnote ²⁶³

Strikingly, User Rights decision-makers have directly applied the CFR to online platforms, adopting what is, at present, a broad interpretation of the ECJ jurisprudence. According to User Rights, Article 14(4) DSA merely formalizes a pre-existing obligation for platforms to enforce fundamental rights.Footnote ²⁶⁴ This approach could certainly merit a standalone analysis.Footnote ²⁶⁵ However, I limit myself here to the more cautious observation that the direct application of EU fundamental rights to social media platforms appears rooted in a noble intent – though its long-term efficacy remains uncertain, given the nonbinding nature of the decisions mandating such application.

When questioned about the nonbinding status of their decisions, one of the cofounders of User Rights emphasized two points. First, he saw one primary challenge to “introduce public law” to platforms’ managerial ecosystem.Footnote ²⁶⁶ While not explicitly stated – though discussed in Chapter 2 – it is understood that public law currently only plays a minimal role in platforms’ internal content moderation due to their global scale. ODSs, then, may represent the first opportunity to compel platforms to engage with local law on a case-by-case basis.Footnote ²⁶⁷ Second, to users, the interviewed cofounder characterizes their role as providing a “right to be heard, akin to a court” and delivering a comprehensive legal analysis.Footnote ²⁶⁸ Once established, the cofounder argues, a well-reasoned, rigorous legal assessment would be difficult for platforms to disregard.Footnote ²⁶⁹ For instance, given that Article 21 DSA is one of the few provisions offering users substantively new options, the User Rights cofounder reported that at numerous conversations in Brussels, the European Commission reiterated its commitment to “make Article 21 DSA work.”Footnote ²⁷⁰

Bringing public law principles into private, managerial settings is no small feat. As discussed in this chapter, ODSs have limited power to compel platforms to act against their preferences – a reality underscored by platforms’ demonstrated resilience in deflecting external pressures throughout the legislative process.

Yet, there are still grounds for cautious optimism. Well-reasoned decisions and ongoing monitoring by the European Commission could foster an ecosystem in which platforms find it increasingly difficult to ignore the external feedback provided by ODSs like User Rights. This feedback’s impact may be amplified by the persuasiveness of legal argumentation, pressure from the Commission, and the legitimizing role of external experts. Just as the Oversight Board leverages the authority of its high-profile members – including a former Head of State, a Nobel Peace Prize laureate, and law professors, which we will examine in the next chapter – User Rights has established an “Advisory Board” on a smaller but similarly strategic scale. Currently, this Advisory Board comprises several renowned young academics from across European jurisdictions (the Netherlands, Croatia, Portugal, and Germany) who are tasked with advising User Rights on complex issues and exploring the future role of ODSs.

3.8.1.3 Appeals Centre Europe

The third player then is the Appeals Centre Europe (the “Appeals Centre”), which set up shop with twenty-five employees in Dublin. It received a whopping fifteen-million-dollar kick off funding by the Oversight Board Trust.Footnote ²⁷¹ That trust is behind what we know as “the Oversight Board” and about which we will learn more in the next chapter.Footnote ²⁷² The Appeals Centre is headed by Thomas Hughes, the founding Director of the Oversight Board and former Executive Director of Article 19, a globally active free speech advocacy group named after the respective provision in the International Covenant on Civil and Political Rights.Footnote ²⁷³

Unlike ADROIT, which focuses more on commercial disputes, and User Rights, which focuses on EU fundamental rights, the Appeals Centre seeks to advance structural reform. In that it is clearly a brainchild of the experiences at the Oversight Board. According to an early interim webpage, holding the place until the official website is launched and the Appeals Centre begins accepting cases, the Centre posits that it “sees the potential for out-of-court dispute settlement bodies to contribute to ensuring greater accountability and transparency of social media companies, through helping to identify recurring areas of concern and systemic risk arising from how their platforms function.”Footnote ²⁷⁴ That is noteworthy for several reasons.

First, compared to the other and at this point smaller ODSs, the Appeals Centre’s focus seems to shift a little from settling individual disputes to using those disputes as leverage to remedy larger, more structural governance flaws at platforms. In an interview, the Director of the Appeals Centre explains that the thinking was, from the start, to operationalize individual cases as structural remedies.Footnote ²⁷⁵ The Director envisions the Appeals Centre as a “heat map” that can identify particularly pressing issues in content moderation.Footnote ²⁷⁶ The Appeals Centre adjudicates very differently to the Oversight Board. The latter decides only a few cases per year and gives structural governance advice. It is, in many respects, not at all about solving disputes. The Appeals Centre, in turn, is about settling as many disputes as possible and, as a side effect, getting a better understanding what the structural problems are. When asked about the relationship between the Appeals Centre and the Oversight Board, Hughes smiled and says “although I’m really not a fan of the Supreme Court analogy when it comes to the Oversight Board, in relation to us it has something to it. The Oversight Board focuses on a few important cases and gives guidance. We, in turn, are a small claims court in this analogy.”Footnote ²⁷⁷

Second, it connects Article 21 DSA to “systemic risk assessments” and “independent audits” as foreseen in Articles 33–37 DSA, a section of the DSA that never came up in interviews with EU lawmakers as connected to ODSs.Footnote ²⁷⁸ The idea is simple. The DSA requires that very large online platforms – actors like Facebook and X – self-assess the “systemic risks in the Union stemming from the design or functioning of their service,” Article 34(1) DSA. Such systemic risks might include the “dissemination of illegal content” or “any actual or foreseeable negative effects on civic discourse and electoral processes, and public security,” Article 37(2)(a) and (c) DSA. In principle, these platforms conduct these reports themselves. Yet, given that they have little incentive to declare their services excessively risky to the societies they operate in, such self-assessments may not be particularly rigorous. External and independent actors could be more reliable stewards of systemic risk assessment. For this reason, the DSA mandates that very large online platforms submit to independent audits once a year.

However, to readers who, like the author, have some – even if only modest – experience working in corporate law, these audits might not seem especially promising on closer inspection. Auditors are tasked with assessing whether Very Large Online Platforms (VLOPs) comply with obligations to assess and mitigate systemic risk, but these are procedural rather than substantive obligations. It would hardly be surprising if VLOPs found it fairly easy to present assessments and mitigation efforts as compliant, while leaving their core practices untouched.

Enter, the Appeals Centre. Given its “small claims court”-like intake of potentially tens of thousands of cases in short periods of time, the Appeals Centre might very well be in a position to “identify, analyse, and assess” systemic risks posed by large online platforms. The Appeals Centre appears, therefore, to already work on ways to manage the “data,” as its interviewed Director put it, of all the cases it handles to aggregate and translate them into modes of action eventually fostering structural governance reforms at platforms.Footnote ²⁷⁹

The appeal of this approach is clear. Unlike the systemic risk assessments conducted by platforms or the independent audits by auditors – an industry with well-known issues across sectors – the Appeals Centre’s assessments of ‘systemic risks’ are grounded in real complaints from actual users. Moreover, platforms are required to fund the Appeals Centre while users pick the ODS, which mitigates a structural moral hazard familiar from rating agencies and auditors, where leniency may be motivated by fear of losing lucrative clients.Footnote ²⁸⁰ With a nominal five-euro fee charged to users to deter frivolous appeals, we can reasonably assume that most of these appeals will highlight genuine grievances, which, as noted in Chapter 2, often reveal underlying structural governance issues.

3.8.2.1 Automated Justice and Large Language Models

First, we are already witnessing productive experimentation within the field of ODSs. Remarkably, the most groundbreaking initiatives appear driven by a blend of profit motives and efforts to operationalize public law principles vis-à-vis platforms. To achieve profitability, ODSs must operate on a substantial scale, processing large volumes of texts and images. Consequently – and this, arguably, was unforeseeable during the drafting of the DSA prior to the launch of ChatGPT – most, if not all, ODSs will likely rely extensively on “artificial intelligence,” especially large language models, to streamline operations.

ODSs aim to establish decisional systems that can manage high case volumes, apply normative frameworks, and – following some degree of human review – issue reasoned decisions or, to be precise and depending on the grade of automation, produce texts that read like reasoned decisions.Footnote ²⁸¹ Some systems are oriented more toward business considerations, others toward fundamental rights, and others still with a direct focus on structural governance. Yet, these are differences of emphasis rather than of fundamental approach. By combining normativity with technological sophistication, ODSs strive to construct what neither content moderation (limited by its profit motive) nor conventional courts (limited by lack of technological expertise) could achieve in the past decade: an efficient, moderately profitable system that processes large quantities of text, applies normative standards, and produces consistent, coherent decisions in individual cases. We are currently in the very early stages of understanding the limitations of integrating automated systems into justice frameworks – and the warnings are plentiful. However, should the bodies established under Article 21 DSA succeed in effectively embedding AI into their services, the potential insights gained could be transformative.

Therefore, in essence, ODSs represent large-scale experiments on integrating AI into content moderation, administrative justice, and judicial proceedings. As such, ODSs hold significant implications for both private and public administrators and adjudicators, as Chapter 2 theorizes that distinctions between these activities are largely matters of degree. While these domains might (and should) follow somewhat distinct norms and procedures, this book demonstrates that sustaining public values will inevitably depend on private administration and enforcement.

All ODSs seem committed to incorporating some degree of automation into their workflows, from automating case handling systems to elements of case assessment and potentially generating written outputs. However, they uniformly assert that the “decision,” as required by Article 21 DSA, will ultimately be made by a human.Footnote ²⁸² Whether this commitment is realistic remains an open question, largely hinging on scale: if each individual decision takes approximately two hours, like ADROIT anticipates it in its Fee Model, ODSs would need hundreds, potentially thousands, of decision-makers.Footnote ²⁸³ This would eventually turn ODSs into operations reminiscent of the content moderation mills of earlier years – a distinctly undesirable prospect. Clearly, automation has a significant role here, and ODSs are poised to advance these solutions. How to keep this process transparent, identify reliable indicators of success and failure, and determine precisely when a human truly makes the final decision are all questions that remain open – and these issues extend beyond ODSs to the broader use of “artificial intelligence” as a whole.

As one cofounder of User Rights noted, they are “experimenting with how to integrate large language models (LLMs) or, eventually, agentic AI into case analysis, initial normative assessments, and text production.”Footnote ²⁸⁴ Similarly, the Director of the Appeals Centre observed that “LLMs are ideally suited to this type of work” and mentioned potential uses for “triage,” facilitating the further processing and escalation of cases.Footnote ²⁸⁵ Setting up and maintaining an LLM is costly; in this respect, the fifteen million in initial funding provided by the Oversight Board Trust gave the Appeals Centre a considerable advantage. When questioned on whether the Appeals Centre would use Meta’s LLMs, the Director clarified that, for the time being, they intended not to use models developed or operated by Meta.Footnote ²⁸⁶ Designing LLMs that truly align with public law values, if possible at all, likely calls for building them from the ground up rather than adapting potentially normatively compromised existing models.

Should some ODSs – each organized as private entities – succeed in unraveling the Gordian knot discussed in Chapter 2, namely, the perceived tension between public-minded decision-making and efficiency on one hand, and profitability on the other, they could set a compelling example for scalable justice systems on a larger scale.

3.8.2.2 Compliance Ecosystem

Another intriguing aspect – both an opportunity and a challenge – emerging from the diversity of ODSs is their potential to broaden the range of compliance mechanisms. So far, we have seen how legislative wrangling eroded many initially intended compliance tools for Article 21 DSA, such as binding decisions, punitive costs, and possibly even naming and shaming. Yet as the varied landscape of ODSs develops, we may witness the beginnings of a complex, multilayered compliance ecosystem.

In other words, the normative force of ODS decisions – despite their lack of classic mediation roles and the minimal direct exchange between platforms and individuals, even within ODSs – might gradually crystallize through a mix of approaches. Although these decisions are nonbinding, the European Commission seems prepared – and reportedly has conveyed as much to ODS founders, according to several interviewees – to support the adjudicative outcomes of ODSs.Footnote ²⁸⁷ While the Commission cannot strictly enforce individual ODS decisions, it could demand explanations for instances of structural or even individual noncompliance by platforms, creating substantial compliance pressure despite the absence of legal obligation.

Second, ODSs might enlist support from academics and activists to examine and advocate for implementation, thereby applying social and discursive pressure for adherence to ODS decisions. Third, we are likely to see trends of specialization, competition, and convergence among the various ODSs. Some may focus on national contexts, such as the Hungarian “Online Platform Vitarendezö Tanács.” Others, like ADROIT, might veer away from content moderation entirely, focusing on other types of issues. Some may prioritize fundamental rights but reflect national predispositions due to language or legal traditions. There could even be ODSs dedicated to specific platforms or issues, like an ODS for art-related disputes or one for LGBTQI issues. Such specializations could enhance expertise and foster identifiable, “manageable” publics that, through intermediaries, may be motivated to demand compliance.Footnote ²⁸⁸

This diversity, however, also presents challenges. We may see vastly different normative interpretations and approaches, along with ODSs potentially influenced by vexatious litigants or politically motivated groups, each pushing content moderation in divergent directions. The influence of coordinated efforts by special interest groups should not be underestimated – as will be discussed in the next chapter.

3.8.2.3 Normative Fragmentation and the Autonomy of EU Law

This diversity poses a fundamental challenge to the prevailing notions of unity and autonomy in EU law. The EU – and particularly the ECJ – may find itself in a “genie out of the bottle” situation. ODSs are now adjudicating fundamental rights, albeit in varied and, over time, inevitably centrifugal ways. This will generate a growing body of (formally nonbinding) decisions, interpretations, and doctrines that apply EU law – especially the CFR – in a patchwork of interpretations.

This isn’t necessarily a problem. From the perspectives of legal pluralism, federal diversity, and specialized expertise, it could even be considered beneficial. After all, one aim of localizing DSA enforcement is to align review processes more closely with the social norms of specific communities (how feasible this is remains to be seen, especially given the influence of large language models, which rely on foundation models that are anything but local).

Yet, from a contemporary standpoint on the autonomy of EU law, the situation seems peculiar.Footnote ²⁸⁹ The EU, typically committed to eradicating divergent interpretations of core principles, has created quasi-adjudicative bodies tasked with applying EU law – yet without any supervisory mechanism over their decisions. Under current doctrine, none of these bodies can refer questions to the ECJ via the preliminary reference procedure, as none qualifies as a “court or tribunal of a Member State.” And why should they? A traditionalist might argue that ODSs are not courts, and their “decisions” are not legally binding; some might not even consider their decisions as law. But this view would be shortsighted. Out-of-court dispute settlements bodies are, by design, normative actors. They are tasked with applying the law and protecting fundamental rights. They are meant to be the first, fastest, and most knowledgeable venue for fundamental rights protection in European content moderation.

ODS decisions will inevitably matter, even if they only contribute to a thick layer of conflicting interpretations, expectations, and degrees of compliance. This accumulation of diverse interpretations will undeniably shape the practical reality of EU law and, particularly, the CFR in the online sphere. The pressing question now is whether the EU – and especially the ECJ – will devise a mechanism to integrate these diverse, imperfect, yet inventive guardians of EU fundamental rights dynamically into the Union’s legal system.