Europe's AI Act gets real
Add Axios as your preferred source to
see more of our stories on Google.
Illustration: Annelise Capossela/Axios
Europe's landmark AI law has moved from an idea to enforcement.
Why it matters: Brussels will soon learn whether it can turn its sweeping AI rulebook into regulation that companies can actually follow—and if it can become a global benchmark for AI standards.
Driving the news: A major enforcement deadline arrived Aug. 2.
- Most notably, new transparency and disclosure requirements are in effect for chatbots and AI-generated content.
- Europe's AI Office has not gone after any companies covered under the law for misconduct yet, but it can now request information from companies or ask for access to models.
What they're saying: "This is a messy piece of legislation," Patrick Van Eecke, a partner at Cooley in Brussels who advises companies on AI Act compliance, told Axios in an interview.
- He said Aug. 2 marked the point when ordinary users may start to encounter the impacts of the act through being notified they are seeing AI-generated material.
The big picture: The AI Act could become a global standard without being a copy-and-paste rulebook for companies. Execution among companies might get messy, but wherever compliance ends up will be influential as companies decide how to operate around the world.
- "We need a standard. This technological moment is happening really quickly. And right now, at least the EU AI Act gives us a standard," Amy Worley, managing director and data protection officer at Berkeley Research Group, told Axios.
- Transparency notices required under the AI Act have already spread, she said. Watermarking can also serve as a litigation defense for companies, she said.
Case in point: Anthropic announced that future Claude models will generate text that contains a watermark to comply with the EU AI Act.
- Anthropic says watermarked text will be indistinguishable to readers, but that it plans to offer a "watermark detection API" to users soon.
- Google and Meta both said in July they would adopt and develop transparency and watermarking tools. Meta said it would work with the EU AI Office to keep their approach to compliance "practical, interoperable and genuinely useful."
- OpenAI is publishing training data summaries in compliance with the Act, it said, along with embedded "provenance signals." Microsoft's compliance with the Act involves internal governance and risk management changes.
What's next: The transparency requirements that took effect in August are only one part of the AI Act's rollout.
- Rules for major categories of high-risk AI systems, such as uses around education, biometrics and migration, along with AI uses in physical products, come later — in Dec. 2027 and Aug. 2028.
- That means it will be years before the full impact of the law can be assessed.
The bottom line: Europe may not get the rest of the world to copy its AI law exactly. But it may not need to.
- If companies start treating the AI Act as the default benchmark for transparency and risk management, Brussels will have shaped global AI rules even where its law doesn't directly apply.
