When regulation meets reality: Does the AI act actually meet its promises?

The timing of the conference could hardly be more relevant. The adoption of the EU's Artificial Intelligence Act (AIA) is widely seen as a defining moment in the legal and regulatory landscape for AI, the first of its kind to attempt a comprehensive legal framework for AI. Its ambition is to make sure AI systems are safe, trustworthy, and respectful of fundamental rights. But turning that ambition into workable practice is far from simple. Behind the AIA's broad principles lie complex legal, ethical, governmental and societal questions. This year's conference theme, "Critical Perspectives of the AI Act," was designed to confront exactly those questions head-on, rather than treating the legislation as a finished product.  

AI regulation develops quickly

Krämer tells us that the implications of AI and its regulations are developing quickly. “Only a couple of weeks ago, on 2 August 2026, the AI Act's transparency obligations became applicable and are now being enforced. From that date, AI generated content must contain a clear disclosure indicating its origin,  meaning that AI generated audio, image, video and text have to carry machine-readable marking. For instance, the LLM Claude by Anthrophic has gathered a lot of attention since they announced that they will implement watermarking in their responses. Therefore, it should be possible to distinguish written texts by Claude soon, even though Anthrophic claims that this will not be visible or distinguishable for readers of the texts.” 

What are the hurdles for good AI regulation? 

The AI Act is designed similarly to product safety laws. Krämer explains why this might be an issue: “While this approach is great for spotting technical errors in products, it might be difficult to address structural problems like discrimination or the loss of independence that many AI systems pose as challenges. ” 

Second, one presentation focused on AI companion bots, and the risks they pose for society. ”Companion bots refer to apps that simulate companionship, for instance by keeping a conversation with its users. They often use mechanisms that make you addicted, or move the conversations to explicit content, which has severe implications especially for minors. While the AI Act does apply to such companies by applying to non-EU actors whose systems are placed in the EU, the question is also about enforcement: Because this company is based in the US, it will be interesting to what extent the AI Act can meaningfully be enforced.”  

Diversity of perspectives and expertise

The focus on this single piece of legislation sparked many different discussions: from deepfakes, dark patterns and regulatory sandboxes to agricultural AI, AI companions, and AI toys. During the conference, researchers were able to learn from each other and make connections across these different topics. “It was great to see the diversity of thought that emerged. This showed clearly that understanding and regulating AI could and should bring together different perspectives and areas of expertise,” Krämer reflects. 

Takeaways for AI legislation development

What ties these discussions together is a shared recognition that the AI Act isn't just a technical or legal document. It's a living framework that will shape, and be shaped by ongoing developments in technology, institutions, and society. Conferences like LAIR matter precisely because they create space for that kind of critical, interdisciplinary discussion, the kind that's essential if AI regulation is going to work not just on paper, but in practice. “We think it is important to have critical scholarship investigating whether the AI Act actually meets its promises, and if it can contribute to its goal of protecting fundamental rights and the promotion of trustworthy AI. LAIR takes one step into this direction by creating a community that does this”, Krämer concludes.