EU AI Act turns enforceable today: apps must disclose and label AI
The AI Office can now fine model makers up to €15M or 3% of revenue; chatbots must out themselves and synthetic media needs machine-readable marks.
Copy markdown
The switch flips today
As of Aug 2, the AI Office and national authorities can enforce everything already in force — prohibited practices, general-purpose-AI (GPAI) obligations, Article 50 transparency, and AI-literacy duties. The rules existed before; the teeth didn't.
Ship generative AI? You have new homework
Article 50 now bites: your chatbot must tell users they're talking to an AI, and any synthetic image, audio, video, or text must carry a machine-readable "AI-generated" mark. The disclosure has to be clear at first interaction, not buried in a footer.
Model-makers face real fines now
The Commission can fine GPAI providers up to €15M or 3% of global turnover — for breaches, ignored documentation requests, or refusing model access for evaluation. Models placed on the market before Aug 2, 2025 get until Aug 2, 2027 to comply.
The exemptions that save you work
Not everything needs a watermark: grammar-style assistive editing, obviously artistic or fantastical content, and AI text under genuine human editorial review are all carved out. Learn them before you bolt a label onto every output you generate.
What's not enforced yet
The heavy high-risk-system rules under Annex III slipped to Dec 2, 2027, so today's line is about transparency and GPAI — not full conformity assessments. Don't over-build compliance you don't owe for another 16 months.