Key transparency and disclosure provisions of the European Union’s AI Act took effect on Aug. 2, 2024, marking the transition from a planned regulatory framework to enforceable obligations for companies operating under the law.
What changed and why it matters
The new rules primarily require notifications and disclosures for chatbots and AI-generated content, meaning ordinary users may start receiving explicit notices that what they see was produced by an AI system. In parallel, the EU AI Office now has statutory authority to request information from companies covered by the Act and to seek access to models for examination. As of the August enforcement deadline, the office has not yet initiated enforcement actions against covered companies.
Expert perspectives
Patrick Van Eecke, partner at Cooley’s Brussels office who advises firms on AI Act compliance, told Axios the legislation is "a messy piece of legislation." He said Aug. 2 is when everyday users are likely to begin seeing the Act’s direct effects through transparency notices.
Amy Worley, managing director and data protection officer at Berkeley Research Group, argued that a standard is necessary given the rapid pace of the technology. "At least the EU AI Act gives us a standard," she told Axios. Worley also noted that transparency notices are already spreading and that watermarking can serve as a litigation defense for companies.
Industry responses — concrete examples
- Anthropic announced that future Claude models will produce text containing a watermark to comply with the EU AI Act. The company says the watermarked text will be indistinguishable to readers but plans to offer a "watermark detection API" to users soon.
- Google and Meta both said in July they would adopt or develop transparency and watermarking tools.
- Meta said it will work with the EU AI Office to ensure their compliance approach is "practical, interoperable and genuinely useful."
- OpenAI said it is publishing training-data summaries in compliance with the Act and embedding provenance signals in outputs.
- Microsoft has been adjusting internal governance and risk-management processes as part of its approach to compliance.
What’s next
The transparency requirements that became effective in August represent only the first phase of the AI Act’s rollout. Rules for major categories of high‑risk AI systems — including applications in education, biometric identification, migration management, and AI embedded in physical products — are scheduled to take effect later, in December 2027 and August 2028. As a result, the law’s full impact will take years to assess.
Bottom line
The EU may not succeed in getting every jurisdiction to replicate the AI Act verbatim, but it may not need to. If companies treat the Act as the de facto benchmark for transparency and risk management, Brussels could shape global AI practices even where its law does not have direct legal force.
