EU AI Act enforcement begins with transparency requirements

Europe's AI Act crossed a critical threshold on August 2 when its first major enforcement provisions took effect, requiring new transparency and disclosure standards for chatbots and AI-generated content.

The European Union's AI Office can now request information from companies and access AI models, though it has not yet pursued any enforcement actions. The milestone marks the transition from legislative text to real-world compliance challenges that will test whether Brussels can make its ambitious regulatory framework practical.

Why it matters

The EU AI Act may establish de facto global standards for AI transparency and risk management even without direct adoption by other jurisdictions. If major technology companies treat European requirements as their baseline for operations worldwide, Brussels will have shaped international AI governance regardless of whether other governments copy the legislation.

Companies respond with watermarking and disclosures

Anthropic announced that future Claude models will generate text containing watermarks to comply with the regulation. The company plans to release a watermark detection API while keeping the marks invisible to readers.

Google and Meta both committed in July to developing transparency and watermarking capabilities. Meta stated it would collaborate with the EU AI Office to ensure its compliance approach remains "practical, interoperable and genuinely useful."

OpenAI is publishing training data summaries and embedding provenance signals in its outputs. Microsoft has implemented internal governance and risk management changes to meet the Act's requirements.

Execution challenges ahead

"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. He noted that August 2 represented the point when ordinary users would begin encountering the law's effects through notifications about AI-generated material.

Amy Worley, managing director and data protection officer at Berkeley Research Group, told Axios that transparency notices required under the Act have already begun spreading beyond Europe. She noted that watermarking can serve as a litigation defense for companies.

Phased rollout extends to 2028

The transparency requirements that took effect in August represent only the first phase of implementation. Rules governing high-risk AI systems in education, biometrics, migration, and physical products will arrive in December 2027 and August 2028.

This extended timeline means the law's full impact cannot be assessed for several years. The staggered approach gives companies time to adapt but also creates uncertainty about compliance standards that remain under development.

Worley emphasized the value of having any standard during rapid technological change: "We need a standard. This technological moment is happening really quickly. And right now, at least the EU AI Act gives us a standard."

These details were first reported by Axios.