This paper investigates the feasibility and potential role of using Large Language Models (LLMs) to support systemic risk audits under the European Union’s Digital Services Act (DSA). It examines how automated tools can enhance the work of DSA auditors and other ecosystem actors by enabling scalable, explainable, and legally grounded content analysis. An interdisciplinary expert workshop with twelve participants from legal, technical, and social science backgrounds explored prompting strategies for LLM-assisted auditing. Thematic analysis of the sessions identified key challenges and design considerations, including prompt engineering, model interpretability, legal alignment, and user empowerment. Findings highlight the potential of LLMs to improve annotation workflows and expand audit scale, while underscoring the continued importance of human oversight, iterative testing, and cross-disciplinary collaboration. This study offers practical insights for integrating AI tools into auditing processes and contributes to emerging methodologies for operationalizing systemic risk evaluations under the DSA.

More people voted in 2024 than any other year in human history, while often relying on the internet for political information. This combination resulted in critical challenges for democracy. To address these concerns, we designed an exhibition that applied interactive experiences to help visitors understand the impact of digitization on democracy. This late-breaking work addresses the research questions: 1) What do participants, exposed to playful interventions, think about these topics? and 2) How do people estimate their skills and knowledge about countering misinformation? We collected data in 5 countries through showcases held within weeks of relevant 2024 elections. During visits, participants completed a survey detailing their experiences and emotional responses. Participants expressed high levels of self-confidence regarding the detection of misinformation and spotting AI-generated content. This paper contributes to addressing digital literacy needs by fostering engaging interactions with AI and politically relevant issues surrounding campaigning and misinformation.

The rise of digital platforms has transformed political campaigning, introducing complex regulatory challenges. This paper presents a comprehensive taxonomy for analyzing political content in the EU’s digital electoral landscape, aligning with the requirements set forth in new regulations, such as the Digital Services Act (DSA). Using a legal doctrinal methodology, we construct a detailed codebook that enables systematic content analysis across user-generated and political ad content to assess compliance with regulatory mandates.

The AI Act represents a significant legislative effort by the European Union to govern the use of AI systems according to different risk-related classes, imposing different degrees of compliance obligations to users and providers of AI systems. However, it is often critiqued due to the lack of general public comprehension and effectiveness regarding the classification of AI systems to the corresponding risk classes. To mitigate these shortcomings, we propose a Decision-Tree-based framework aimed at increasing legal compliance and classification clarity. By performing a quantitative evaluation, we show that our framework is especially beneficial to individuals without a legal background, allowing them to enhance the accuracy and speed of AI system classification according to the AI Act. The qualitative study results show that the framework is helpful to all participants, allowing them to justify intuitively made decisions and making the classification process clearer.

Content moderation is a vital condition that online platforms must facilitate, according to the law, to create suitable online environments for their users. By the law, we mean national or European laws that require the removal of content by online platforms, such as EU Regulation 2021/784, which addresses the dissemination of terrorist content online. Content moderation required by these national or European laws, summarised here as ‘the law’, is different from the moderation of pieces of content that is not directly required by law but instead is conducted voluntarily by the platforms. New regulatory requests create an additional layer of complexity of legal grounds for the moderation of content and are relevant to platforms’ daily decisions. The decisions made are either grounded in reasons stemming from different sources of law, such as international or national provisions, or can be based on contractual grounds, such as the platform's Terms of Service and Community Standards. However, how to empirically measure these essential aspects of content moderation remains unclear. Therefore, we ask the following research question: How do online platforms interpret the law when they moderate online content? To understand this complex interplay and empirically test the quality of a platform's content moderation claims, this article develops a methodology that facilitates empirical evidence of the individual decisions taken per piece of content while highlighting the subjective element of content classification by human moderators. We then apply this methodology to a single empirical case, an anonymous medium-sized German platform that provided us access to their content moderation decisions. With more knowledge of how platforms interpret the law, we can better understand the complex nature of content moderation, its regulation and compliance practices, and to what degree legal moderation might differ from moderation due to contractual reasons in dimensions such as the need for context, information, and time. Our results show considerable divergence between the platform's interpretation of the law and ours. We believe that a significant number of platform legal interpretations are incorrect due to divergent interpretations of the law and that platforms are removing legal content that they falsely believe to be illegal (‘overblocking’) while simultaneously not moderating illegal content (‘underblocking’). In conclusion, we provide recommendations for content moderation system design that takes (legal) human content moderation into account and creates new methodological ways to test its quality and effect on speech in online platforms.

Large Language Models (LLMs) are expected to significantly impact various socio-technical systems, offering transformative possibilities for improved interaction between humans and technology. However, their integration poses complex challenges due to the intricate interplay between societal structures, human behaviour, and technological innovation. This research explores these multifaceted challenges, emphasising the need for a human-centered approach in integrating LLMs to ensure that technological advancements are aligned with ethical standards and societal needs. Utilizing a structured methodology comprising a workshop, literature analysis, and expert collaborations, the study uses a multi-dimensional human-centered AI framework to guide the responsible integration of LLMs. Key insights include the importance of inclusive data, considering unintended consequences, maintaining privacy, and respecting intellectual property rights. The paper identifies and advocates for principles like human-in-the-loop, continuous longitudinal studies, proactive awareness campaigns, and regular audits to develop LLMs that are ethically sound, adaptable, and effectively integrated into various socio-technical systems, thus addressing user needs and broader societal impacts. The paper also underlines the importance of collaboration among academia, industry, and policymakers to develop LLMs that are ethically aligned, socially beneficial, and adaptable to future societal needs. The findings offer valuable insights into the strategic integration of LLMs, advocating for a broader research perspective beyond industrial motivations to fully understand and leverage LLMs in socio-technical landscapes.

Content moderation is a vital condition that online platforms must facilitate according to the law and to create (adequate) online environments for their users. While new regulatory requests, like the Digital Services Act in the European Union create novel obligations for platforms, other legal dimensions like the law of Member States are an additional layer of legal grounds for the moderation of content and relevant for the decisions taken on a day-by-day basis. The decisions taken are either grounded in reasons stemming from law or can be based on contractual grounds like the platform’s Terms of Service and their Community Standards. How to measure these essential aspects of content moderation empirically however is still unclear. We therefore ask the following research question: How do online platforms interpret the law when they moderate online content?To understand this complex interplay and to test the quality of the platform’s content moderation claims empirically, this article develops a methodology that facilitates empirical evidence on the individual decisions taken per piece of content, while highlighting the subjective element of content classification by human moderators. We then apply this methodology to a single empirical case, an anonymous medium-sized German platform which has provided us access to their content moderation decisions. By better understanding how platforms interpret the law we can understand how complex content moderation, its regulation and compliance practices are, as well as to what degree legal moderation might differ from moderation due to contractual reasons in dimensions like the need for context, information, or time.Our results show a considerable divergence between the platforms interpretation of the law and our own interpretation. We believe that a significant number of platform legal interpretations are incorrect. These divergent interpretations of the law mean that we believe platforms are removing legal content that they falsely believe to be illegal (‘over-blocking’) while simultaneously not moderating illegal content (‘under-blocking’). In conclusions, we provide recommendations for content moderation system design that takes (legal) human content moderation into account and creates new methodological ways to test their quality and effect on speech on online platforms.

Safeguarding democratic elections is hard. Social media plays a vital role in the discourse around elections and during electoral campaigns. The following article provides an analysis of the ‘systemic electoral risks’ created by Twitter and Facebook and the mitigation strategies employed by the platforms. It is based on the 2020 proposal by the European Commission for the new Digital Services Act (DSA) in the context of the 2021 German federal elections. This article focuses on Twitter and Facebook and their roles during the German federal elections that took place on 26 September 2021. We analysed three systemic electoral risk categories: 1) the dissemination of illegal content, 2) negative effects on electoral rights, and 3) the influence of disinformation and developed systematic categories for this purpose. In conclusion, we discuss how to respond to these challenges as well as avenues for future research.

Safeguarding democratic elections is hard. Social media plays a vital role in the discourse around elections and during electoral campaigns. The following article provides an analysis of the 'systemic electoral risks' created by Twitter and Facebook and the mitigation strategies employed by the platforms. It is based on the 2020 proposal by the European Commission for the new Digital Services Act (DSA) in the context of the 2021 German federal elections. This article focuses on Twitter and Facebook and their roles during the German federal elections that took place on 26 September 2021. We analysed three systemic electoral risk categories: 1) the dissemination of illegal content, 2) negative effects on electoral rights, and 3) the influence of disinformation and developed systematic categories for this purpose. In conclusion, we discuss how to respond to these challenges as well as avenues for future research.

The AI Act represents a significant legislative effort by the European Union to govern the use of AI systems according to different risk-related classes, linking varying degrees of compliance obligations to the system's classification. However, it is often critiqued due to the lack of general public comprehension and effectiveness regarding the classification of AI systems to the corresponding risk classes. To mitigate those shortcomings, we propose a Decision-Tree-based framework aimed at increasing robustness, legal compliance and classification clarity with the Regulation. Quantitative evaluation shows that our framework is especially useful to individuals without a legal background, allowing them to improve considerably the accuracy and significantly reduce the time of case classification.

Basic human rights, like freedom of expression, freedom of the press, and privacy, are being radically transformed by new technologies. The manifestation of these rights in online spaces is known as “digital rights,” which can be impeded or empowered through the design, governance, and litigation of emerging technologies. Design defines how people encounter the digital world. Some design choices can exploit the right to privacy by commodifying attention through tactics that keep users addicted to maximize profitability; similar design mechanisms and vulnerabilities have facilitated the abuse of journalists and human rights advocates across the globe. But design can also empower human rights, providing novel tools of resistance, accountability, and accessibility, as well as the inclusion of previously underserved voices in the development process. The new capabilities offered by these technologies often transcend political boundaries, presenting complex challenges for meaningful governance and regulation. To address these challenges, collaborations like the Internet Governance Forum and NETmundial have brought together stakeholders from governments, nonprofits, industry, and academia, with efforts to address digital rights like universal internet access. Concurrently, economic forces and international trade negotiations can have substantial impacts on digital rights, with attempts to enforce steeper restrictions on intellectual property. Private actors have also fought to ensure their digital rights through litigation. In Europe, landmark cases have reshaped the international management of data and privacy. In India, indefinite shutdowns of the internet by the government were found to be unconstitutional, establishing online accessibility as a fundamental human right, intimately tied with the right to assembly. And in Africa, litigation has helped ensure freedom of speech and of the press, rights that may affect more individualsas digital technologies continue to shape media. These three spheres—design, diplomacy, and law—illustrate the complexity and ongoing debate to define, protect, and communicate digital rights.

It is as dictators at the rostrum that figures such as Teodor Currentzis, Arturo Toscanini, and Herbert von Karajan are frequently described, in contrast to the principle of democratic leadership exemplified by such figures as Bruno Walter, Marie Jacquot, and Johannes Wildner. But must a conductor indeed work differently these days, and have the demands being made of conductors changed?

The internet is a place where the political opinion of voters is ever more formed on platforms and their user-generated content globally. A sphere in which the right to free- dom of expression, information and free and fair election are core human rights normative safeguards for our democracy. Securing this process for whoever is not an easy task, as examples like the 2016 US election, the Brexit campaign or the events of the 6th of Jan- uary 2021 illustrate. The European Union has taken regulatory action to secure the digital manifestations of elections, by issuing legislation like the General Data Protection Regu- lation, the Artificial Intelligence Act, the Digital Services Act (DSA)or the Proposal for a regulation of the European Parliament and the European council on the transparency and targeting of political advertising. The aim is to make platforms more transparent, regarding their algorithms deciding on recommendations, price of the ad, or to standardize content moderation to a certain degree. Platforms on the other hand use their Terms of Service (ToS) to implement their Community Standards – a selection of law-like clauses allowing for deletion or blocking of content – to set quasi-norms to safeguard democracy on- line. The ToS used by very large platforms (VLOP) according to Art 25 DSA however does not include granular clauses for European campaigning. The more recent design solutions on platforms include advertising repositories, or warning labels attached to problematic content to better inform the public. However, moderation of content addressing the heart of democracy and the democratic process per se is crucial for the status of human rights in Europe. The first decision taken on a piece of content – if it should be uploaded on the platform or not – is usually automated and controlled by machine learning algorithms. The system in place selects the pieces of content that will be decided upon in the next process step by a human. The moderation of political speech, however, is not solely text-based but does include a fine line of sarcastic elements, emojis or visual content to express itself which is another obstacle to moderating in an electoral context. This article, therefore, asks the question about how to better safeguard the right to fair elections, the right to freedom of expression and information in online campaigning and elections adhering to the recent European legislation, such as the GDPR, the DSA and the AIA and the proposal on the transparency and targeting of political advertising? The article answers the question by taking a closer look at the publicly available data published by platforms on behalf of their transparency reports. Furthermore, the ToS and Community Standards should be analyzed and compared. The process and architecture of content moderation for the selected online platforms are described and modelled according to the publicly available information. Only by providing a more concrete look at content moderation design and practice better solutions for the digital future of democracy can be crafted.

Content moderation is protecting human rights such as freedom of speech, as well as the right to impart and seek information. Online platforms implement rules to moderate content on their platforms through their Terms of Service (ToS), which provides for the legal grounds to delete content. Content moderation is an example of a socio-technical process. The architecture includes a layer that classifies content according to the ToS, followed by human moderation for selected pieces of content. New regulatory approaches, such as the Digital Services Act (DSA) or the Artificial Intelligence Act (AIA) demand more transparency and explainability for moderation systems and the decisions taken. Therefore, this article answers questions about the socio-technical sphere of human moderation: • How is certainty about content moderation decisions perceived within the moderation process? • How does the measurement of time affect content moderator’s work? • How much context is needed to take a content moderation decision? A sample of 1600 pieces of content was coded according to international and national law, as well as on the Community Standards developed by Meta, mimicking a content moderation scenario that includes lex specialis for content moderation – the German Network.

Technical systems accompany us in our everyday lives, interact with us, influence us and guide our decisions. However, these systems do not only consist of a technical aspect alone, but are much more socio-technically integrated into specific human application contexts. One example of such a socio-technical system is Google Maps.

Who gets to see what on the internet? And who decides why? These are among the most crucial questions regarding online communication spaces – and they especially apply to job advertising online. Targeted advertising on online platforms offers advertisers the chance to deliver ads to carefully selected audiences. Yet, optimizing job ads for relevance also carries risks – from problematic gender stereotyping to potential algorithmic discrimination. The winter 2021 Clinic Increasing Fairness in Targeted Advertising: The Risk of Gender Stereotyping by Job Ad Algorithms examined the ethical implications of targeted advertising, with a view to developing feasible, fairness-oriented solutions. The virtual Clinic brought together twelve fellows from six continents and eight disciplines. During two intense weeks in February 2021, they participated in an interdisciplinary solution-oriented process facilitated by a project team at the Alexander von Humboldt Institute for Internet and Society. The fellows also had the chance to learn from and engage with a number of leading experts on targeted advertising, who joined the Clinic for thought-provoking spark sessions. The objective of the Clinic was to produce actionable outputs that contribute to improving fairness in targeted job advertising. To this end, the fellows developed three sets of guidelines – this resulting document – that cover the whole targeted advertising spectrum. While the guidelines provide concrete recommendations for platform companies and online advertisers, they may also be of interest to policymakers.