Jacob van de Kerkhof

Content moderation is the process of organizing and recommending content on the internet and social media. Intermediary services, including very large online platforms like Facebook, YouTube, and TikTok, determine the tone of online discourse and the effective limits of freedom of expression through content moderation. This dissertation examines how the Digital Services Act ('DSA') provides a network of actors with access to content moderation, and how this affects the way freedom of expression is balanced against other rights and interests. The DSA sets requirements for this process and gives various actors a formal role through which they can influence the content moderation process. This role is assigned to, among others, users, civil society, and state actors. Because each of these actors has its own interests, it is possible that freedom of expression may be compromised. How the DSA shapes the involvement of these actors is therefore of crucial importance and forms the core of this dissertation.

Chapter I forms the introduction to this dissertation. First, the research question is introduced: How does the DSA and the formal role it creates for intermediary services, users, civil society, and state actors affect the weighing of freedom of expression in content moderation? This research question is answered by looking at three relevant groups of actors: intermediary services, users and civil society, and state actors. To answer this, various methods have been used. In addition to pure legal research, empirical research methods from computer and media sciences are employed. This is necessary to describe how content moderation works and how intermediary services comply with the DSA.

Besides the methodology, Chapter I outlines the background of the dissertation. The DSA is a relatively new European regulation creating a harmonized regime for intermediary services, including online platforms and search engines. The DSA succeeds the 2000 e-Commerce Directive. It represents a much-needed legal intervention in the online domain: the internet has changed significantly since 2000, bringing new dangers. A gap exists between these dangers and existing legal instruments. The main goals of the DSA are to prevent fragmentation between member states and encourage responsible behavior by intermediary services to ensure a safe, predictable, and trustworthy online environment for EU citizens. The DSA aims to achieve these goals with a series of due diligence obligations that increase transparency and allow intermediary services to be held accountable for their actions. These obligations follow a layered structure: as the intermediary service grows in size and societal impact, more obligations apply. Most obligations apply to Very Large Online Platforms and Very Large Online Search Engines.

Part of the due diligence obligations regarding content moderation consists of creating a network of actors. This can be seen in light of Robyn Caplan's theory on networked platform governance. She describes how platform governance can be delegated to multiple actors who collectively determine the management of a platform within a network. These actors all have different power positions and levels of influence. This makes it difficult to determine to whom the result of that platform governance—how content is moderated and thus how freedom of expression is weighed—can ultimately be attributed. This tension is investigated in the following chapters.

Chapter II provides an empirical background into the content moderation landscape. The chapter examines how the DSA transparency database provides insight into content moderation practices and whether intermediary services indeed provide transparency. Findings from an empirical analysis of the database show that while the database architecture seems suitable for providing transparency, the implementation by intermediary services sometimes leaves much to be desired. It is striking that the majority of content moderation decisions occur outside of social media platforms. Almost all content moderation decisions are made based on terms and conditions rather than applicable law. Additionally, it is notable that intermediary services use different moderation methods. Some report doing everything automatically, while others state that all moderation is performed by human moderators. The value of the information provided is questionable: in many cases, it is unclear why content was removed because intermediary services use generic references to general terms. The main conclusions of this chapter for the dissertation are that intermediary services largely moderate based on their own terms and conditions, and that the degree of knowledge sharing between actors is largely determined by the extent to which intermediary services provide knowledge.

Chapter III focuses on the liability exemption enjoyed by intermediary services under the DSA. The academic consensus is that this has not changed since the e-Commerce Directive and that the due diligence obligations in the DSA should be seen as separate from the liability exemption. Chapter III argues that this conclusion can be nuanced. Compared to the e-Commerce Directive, a higher degree of diligence can indeed be expected in some cases, which could potentially exclude intermediary services from the liability exemption. A stricter interpretation of the liability exemption could, in some cases, better enable victims of illegal content to hold the intermediary service liable for hosting illegal content. This stricter interpretation could serve as an encouragement to facilitate better notice-and-action mechanisms, sharpen moderation practices against illegal content, and achieve a better balance between the interests of the intermediary service and the injured party. Freedom of expression, as interpreted by the European Court of Human Rights (ECHR), does not necessarily oppose this interpretation. Nor would this interpretation be unsupported by the DSA. Article 16 stipulates that hosting services must offer an easily accessible and user-friendly notice-and-action mechanism. Not offering such a mechanism disrupts the possibilities for holding hosting services liable. To prevent a dysfunctional mechanism from being beneficial, a hosting service that evades the obligations of Article 16 could be excluded from the liability exemption. The same applies to online platforms that structurally evade the obligations of Article 23 DSA. Article 23 obliges online platforms to block users who frequently upload manifestly illegal content. If an online platform fails to do so, it does not act in good faith and can thus be excluded from the Article 6 DSA liability exemption. The chapter also argues that with tightened due diligence obligations regarding manifestly illegal, terrorist, and child abuse content, the bar can be lowered compared to the e-Commerce Directive for factual assessment of whether a hosting service is aware of illegal content and is thus obliged to remove it. This could also lead to the exclusion from the liability exemption of actors acting in bad faith. The chapter argues that intermediary services must thus provide a greater degree of accountability for how they set up their content moderation processes.

Chapter IV focuses on a specific aspect of the liability issue: the notice-and-action mechanism. This mechanism plays a key role in holding hosting services liable for illegal content and allows users to make their preferences known and influence content moderation. The chapter uses an empirical walkthrough to examine how AliExpress, TikTok, Twitch, and X shape their notice-and-action mechanisms, whether this complies with the requirements of Article 16 DSA, and whether this affects the potential liability of hosting services. The research shows that these intermediary services design their notice-and-action mechanisms differently, often tailored to the service they offer. The chapter also shows that there are major differences in easy accessibility between intermediary services. This is because notice-and-action mechanisms for the DSA are often strongly interwoven with mechanisms used for violations of general terms, even though the DSA makes—and requires—a sharp distinction between them. This strong interweaving can also lead users to report content outside the Article 16 mechanism. The chapter concludes that the design of existing mechanisms can be misleading for users. Moreover, the design influences liability for potential illegal content, especially since mechanisms for reporting content under terms and conditions often do not provide the opportunity to offer insight into why certain content is deemed illegal. If users do not provide sufficient explanation for the qualification of content as 'illegal,' hosting services do not have to remove it. The chapter further shows that compliance with Article 16 is varied and that there is room for improvement among several intermediary services. Because notice-and-action mechanisms are an important way for users to make their preferences known—and thus contribute to content moderation and the weighing of freedom of expression—stricter enforcement of Article 16 is necessary.

Chapter V explores Article 22 DSA, which concerns trusted flaggers. Trusted flaggers can be state or non-state entities, such as NGOs, ministries, regulators, or the police. Trusted flaggers have a privileged position in the content moderation process; their notifications must be treated with priority and without delay. Trusted flaggers primarily operate in a bilateral relationship with the online platform, leading to opacity in their actions. Additionally, online platforms likely feel pressure to follow removal requests from trusted flaggers due to their privileged position. This can lead to the removal of legal content, putting pressure on freedom of expression. Due to the opacity of their operations, this can be difficult to correct. Article 22 DSA formalizes this process. Trusted flaggers are appointed by the national digital services coordinator. Their status gives them access to all online platforms—not just those with which they have a bilateral relationship. Article 22 also makes the framework in which trusted flaggers operate more transparent. They may only notify online platforms of illegal content, meaning they must operate within the limits of national and EU legislation. Furthermore, they are expected to report on all their notifications, which counters excessive use of their privileges. This addresses part of the criticism regarding the original trusted flagger framework. Regulation via Article 22 DSA also raises questions. Trusted flaggers are appointed by national authorities and often use national law. The provision contains few safeguards to ensure that national law is not rolled out across the Union, which could also put pressure on freedom of expression. To combat these risks, the chapter advocates for a harmonized portal that trusted flaggers use to report content. Another weak point of Article 22 DSA is that it is unclear what the independence requirement means in the appointment of trusted flaggers. Although intended to prevent financial flows from intermediary services, in practice, trusted flaggers are often financially dependent on them. The chapter suggests that the European Commission could facilitate trusted flaggers receiving fair financial compensation to support their task in countering illegal content.

Chapter VI examines how enforcement by digital services coordinators relates to freedom of expression. Digital services coordinators are national supervisors of DSA compliance. In their supervisory role, they can potentially have a major impact on freedom of expression on the internet; they may sanction content moderation practices, fine intermediary services, and even block access in general. Their role perception thus has a significant influence on the accessibility of certain content and thus on freedom of expression. The enforcement structure for digital services coordinators is complex. They are intended to be independent institutions but are often housed within existing regulators, such as media and consumer authorities. This fragments the supervisory landscape in Europe. This is problematic because the jurisdiction of a digital services coordinator does not extend beyond the intermediary services established within national borders. A complicating factor is that digital services coordinators share their enforcement task with the European Commission, which is authorized to oversee and enforce regarding Very Large Online Platforms and Search Engines. This means cooperation is required. Cooperation in cross-border issues and within the Union raises problems. The first problem is that the illegality of content is determined by both national and EU law. While there is some agreement within the EU on what is illegal, member states' views differ sharply in some areas. The Hungarian anti-LGBTQ+ laws are a striking example. Although the illegality of certain expressions must also be tested against the Charter, it is unclear whether supervisors are capable of making this assessment properly, let alone whether intermediary services do so themselves. This potentially puts freedom of expression under pressure. A second problem is that the jurisdictional structure of the DSA causes enforcement to concentrate in a small number of member states, particularly Ireland and the Netherlands. Many complaints must be handled by the Irish or Dutch supervisor, which can cause operational bottlenecks. Freedom of expression on the internet can come under pressure when national supervisors become overwhelmed. A final complication in the DSA enforcement structure is that national enforcement strategies can vary widely, specifically due to digital services coordinators being housed within supervisors from different legal fields. This applies even more when resources are lacking in smaller member states. This too can put the protection of freedom of expression at risk. The chapter concludes that thought must be given to mechanisms through which supervisors can be forced into action, horizontal preliminary procedures between member states to test interpretations of national law, and redistribution of the supervisory fee from Article 43 DSA to support small member states. In this way, freedom of expression can be less pressured during the enforcement of the DSA by digital services coordinators.

Chapter VII focuses on the interference of state actors in content moderation. When a state actor without formal authority exerts pressure on intermediary services to moderate certain types of content, this can be seen as jawboning. Jawboning is a term used in (predominantly American) literature to describe how an actor tries to achieve a policy goal outside their authority by exercising (informal) pressure on private actors. When this is done to influence content moderation, it clashes with freedom of expression. After all, restrictions on freedom of expression must be justified, meaning they must have a legal basis, pursue a legitimate goal, and be proportional and necessary. This cannot be tested as a whole when the restriction occurs informally. A current example of jawboning in content moderation is guidelines such as the code of conduct against hate speech. In principle, such guidelines and codes of conduct are proposed as forms of self- or co-regulation. However, a legislator often strongly encourages private actors to participate, facilitates negotiations, or even threatens far-reaching legislation to achieve the policy goals underlying the codes. This is also evident with the code of conduct against hate speech. The Commission explicitly indicated that if the code were not sufficiently complied with, stricter legislation could be expected. It steered toward more proactive measures against hate speech. Such an approach clashes with freedom of expression, particularly because the boundaries of hate speech are vague and highly context-sensitive. Encouragement toward more proactive moderation can lead to unnecessary restrictions on users' freedom of expression. This would then be the result of the Commission's actions, which should be tested against the ECHR and the Charter. However, this is difficult because it often involves informal pressure. The chapter therefore calls for transparency regarding state actors' input in co- or self-regulatory processes, so their actions can be better tested.

Chapter VIII brings together the findings of the different chapters. It outlines how the interactions created by the DSA between the various actors involved can create tensions and that it is unclear how freedom of expression is weighed in content moderation processes. In terms of Caplan's theory on shared platform governance, this means that while all actors have influence on content moderation, it remains unclear how this influence relates to the weighing of freedom of expression. Based on the suggestions made in the individual articles, the chapter calls for a strengthening of the possibilities and obligations of individual actors to provide accountability for their role in content moderation processes. It also needs to be clarified how their input influences freedom of expression. Besides these improvements, it is necessary to create clarity about what the DSA can and cannot achieve, thereby ensuring it does not become a Swiss army knife without a legal basis. Only in this way can sustainable implementation and enforcement of the DSA be guaranteed.