EU AI Act now forces chatbots to disclose and AI content to be marked
Live since Aug 2: bots must ID as bots, synthetic media needs machine-readable marks, and non-EU builders serving EU users risk EUR 15M or 3%-of-revenue fines.
Copy markdown
Your chatbot has to say it's a bot
Article 50 now requires any user-facing AI to disclose it isn't human at the first interaction, clearly and accessibly. If you ship a support bot, assistant, or agent front-end, that's a disclosure you legally owe every EU user.
Generated media needs a machine-readable tag
Synthetic audio, images, video, and text must carry machine-readable AI markers, and deepfakes must be labeled. Only output published on or after Aug 2 counts, so now is the moment to wire watermarking or provenance metadata into your generation pipeline.
Fines top EUR 15M or 3% of global revenue
Enforcement went live Aug 2 with public complaint and anonymous whistleblower tools already open. Penalties reach EUR 15M (about $17.3M) or 3% of worldwide turnover, whichever is higher.
Non-EU indie devs are in scope too
The rules bind anyone deploying interactive AI or general-purpose models to EU users, wherever the company sits: a US solo dev shipping a chatbot to European customers is covered. The Commission's voluntary Code of Practice on Transparency, already 180+ signatories, spells out the marking, metadata, and 'AI' label standards to follow.
Elsewhere: OpenAI locks its cyber models to enterprise partners
OpenAI's new Daybreak program (Aug 10) routes its frontier cyber models — Daybreak Blue for defense, Red for red-teaming — only to vetted firms like CrowdStrike, IBM, and Accenture, and says access explicitly won't reach individual customers or indie builders.