The EU AI Act gets real

Brussels has ushered in enforcement for the EU AI Act, marking a shift from theoretical framework to tangible regulations. A key enforcement deadline arrived on August 2, when new transparency and disclosure requirements went into effect for chatbots and AI-generated content. The EU AI Office has not yet penalized any companies for violations, but can now request information and seek access to models.

Patrick Van Eecke, a Brussels-based AI law expert, noted that the deadline signaled when ordinary users could start experiencing the impacts of the law through notifications regarding AI-generated material. The AI Act could become a global standard without being a direct copy for companies, but its implementation might be messy. Transparency requirements, however, are already being adopted by major tech firms, with Anthropic, Google, and Meta expressing their commitment to compliance.

Anthropic plans to include watermarks in future Claude models, which will be indistinguishable to readers but will allow users to access a watermark detection API. Meanwhile, Google and Meta have pledged to develop transparency and watermarking tools, with Meta aiming for practical, interoperable, and genuinely useful solutions.

OpenAI has begun publishing training data summaries and embedded "provenance signals" in line with the Act. Microsoft is focusing on internal governance and risk management changes to comply with the EU AI Act.

While the AI Act is not a one-size-fits-all solution, its influence could still shape global AI rules, even if its law doesn't directly apply. The rollout of the AI Act has multiple phases, with major high-risk AI system rules scheduled for later in 2027 and 2028. Overall, Europe may not dictate the exact copy of its AI law, but its benchmark for transparency and risk management could become influential worldwide.