On Aug. 2, the European Commission’s AI Office began enforcing the AI Act, switching on Article 50’s transparency duties and the general-purpose AI regime under Regulation (EU) 2024/1689. Brussels is now the first jurisdiction on Earth with a live AI regulator, and its jurisdiction runs wherever model outputs reach the single market.

The substantive rules are narrow but sharp. Chatbots, voice assistants, and agentic systems must disclose that they’re AI. Deepfakes require labelling. Generative outputs must carry machine-readable marks. Providers already shipping systems get until Dec. 2, 2026 to retrofit that marking. Per Cooley’s Aug. 3 analysis, noncompliance carries fines up to €15 million or 3% of worldwide annual turnover, whichever is higher.

The AI Office paired the enforcement date with a voluntary Code of Practice on Transparency of AI-Generated Content, adopted alongside the Article 50 guidelines on July 20, complete with a standard icon set. Signatories get a presumption of conformity and a softer touch on enforcement. It’s the same regulatory grammar Brussels used for GDPR codes: bring the industry inside the tent, and let holdouts wear the litigation risk.

Political heat is already gathering. Michael McNamara, MEP (Renew, Ireland), told Euronews the Trump administration may frame enforcement “as an attack on US commercial interests,” pointing to December 2025’s Digital Markets Act fight and this month’s Google DMA fine as the template.

Washington isn’t dormant either. The Transparency Coalition counts 27 states with 85 AI-related laws passed in 2026, and on Aug. 13 California’s Assembly and Senate Appropriations committees run suspense-file votes on roughly 30 AI bills covering chatbot safety, frontier model oversight, and children’s digital lives.

The federal vacuum is being filled from both coasts of the Atlantic. Brussels just moved first.

Sources

Sources