Was stimmt nicht mit dem neuen europäischen Gesetz über künstliche Intelligenz?

In a move that has sparked debate among tech companies, regulators, and civil society groups, the European Commission has classified ChatGPT, Reddit, and Roblox under the stricter provisions of the Digital Services Act (DSA). This classification means these platforms must comply with additional obligations within four months, starting in January 2027. The decision follows reports that each service has exceeded 45 million monthly users within the EU, meeting the threshold set by the DSA to qualify as very large online platforms (VLOPs) or very large search engines (VLOSE). The European Commission announced the classifications on Monday, following notifications from operators confirming their user numbers met the criteria outlined in the legislation governing digital services. These new obligations include assessing and mitigating systemic risks arising from their services and algorithmic systems. Risks encompass the spread of illegal content, negative impacts on minors, physical and mental well-being of users, fundamental rights, electoral processes, and public safety. Operators will need to conduct detailed risk assessments and implement measures to address them, while also regularly reporting on progress. The classification grants the European Commission expanded oversight powers, allowing it to evaluate functions underlying the services and related systems. According to media reports, the commission will collaborate with national authorities in Ireland and the Netherlands, where the relevant service providers are registered, to ensure compliance with the regulation. So far, the commission has designated 28 very large online search engines and platforms under the DSA. Critics have raised several concerns regarding the new regulatory framework. One major point of contention is the legal stretching of the law and misclassification of services. Legal experts argue that ChatGPT and similar systems do not neatly fit into any category defined by the DSA, which traditionally covers hosting, caching, and search functionalities. Classifying a chatbot as a search engine creates legal uncertainty and sets a precedent that could affect other AI-based services. Another concern is double regulation, where providers of large models are already subject to the AI Act. Now they face even stricter obligations under the DSA, including risk assessments, independent audits, and specific procedures for complaints and appeals. This dual regulatory burden raises costs and confusion over which rules apply in different scenarios. Some embedded AI systems on platforms, such as Grok on X, might fall under the DSA alone, potentially leading to inconsistent standards for similar services. Critics also warn of threats to freedom of expression and potential censorship by authority. The DSA requires very large providers to assess and mitigate "systemic risks," including those affecting social discourse and electoral processes. The broad and undefined nature of this term is seen as problematic, potentially forcing providers to remove legitimate content due to excessive caution or enabling political pressure on private companies to limit speech that governments themselves cannot legally prohibit. It could create a de facto truth arbiter deciding what constitutes misinformation or harmful speech. The administrative burden imposed by the DSA is another issue. Requirements such as annual risk assessments, independent audits, transparency of recommendation systems, and dedicated channels for complaints and appeals are particularly demanding. While large American firms can afford these requirements, European and smaller AI providers may struggle, raising fears that companies in the EU might delay or avoid introducing new features altogether due to fear of penalties, which could amount to up to six percent of global traffic. Additionally, critics argue that the tools provided by the DSA are not suitable for addressing specific AI-related risks. Mechanisms such as notification and action, and trusted flaggers, were designed for user-generated content. However, with generative AI, issues such as hallucinations, systematic bias, and the potential for mass creation of misleading content are not adequately addressed by the DSA. Critics suggest that the AI Act would be better suited to handle these challenges. The impact on media and pluralism is yet another area of concern. As the DSA comes into effect, its influence on how information is shared and regulated online is likely to grow, potentially reshaping the digital landscape in ways that could affect free expression, innovation, and competition.