Guide
EU AI Act Article 50, explained in plain English
Article 50 is the transparency backbone of the EU AI Act. It became enforceable on August 2, 2026. Here is what it requires, who it catches, and what happens if you ignore it.
The four obligations
Art. 50(1) — Tell people they're talking to AI
Providers of AI systems that interact directly with people (chatbots, voice agents, AI assistants) must inform users they are dealing with AI — unless it is obvious to a reasonably well-informed person. A support bot that could plausibly pass for a human agent needs a disclosure.
Art. 50(2) — Mark synthetic content in a machine-readable way
Providers of generative AI must ensure outputs (images, audio, video, text) are marked as artificially generated in a machine-readable format. In practice that means watermarks (like SynthID), cryptographic provenance metadata (like C2PA Content Credentials), or both. The marking must be effective, interoperable and robust so far as technically feasible.
Art. 50(3) — Disclose emotion recognition and biometric categorisation
Deployers running emotion recognition or biometric categorisation systems must inform the people exposed to them. This targets workplaces, retail analytics, and similar settings.
Art. 50(4) — Label deepfakes and AI text on matters of public interest
Deployers publishing AI-generated or manipulated content depicting real people, places or events must disclose it visibly. AI-generated text published to inform the public must also be labeled, unless it has undergone human editorial review with a person taking responsibility.
Who is caught (yes, including non-EU companies)
The AI Act applies extraterritorially. If your AI system is placed on the EU market, or its outputs are used in the EU, you are in scope regardless of where your company sits. A US startup whose image generator is used by EU customers has Article 50(2) marking duties. A UK publisher whose AI-written articles reach EU readers has Article 50(4) labeling duties. Run the 2-minute scope check to see which obligations apply to your setup.
The dates that matter now
- August 2, 2026 — Article 50 obligations became enforceable across the EU.
- December 2, 2026 — the grace period for systems placed on the market earlier closes: their synthetic outputs must carry machine-readable marking.
- February 2, 2027 — Commission guidance anticipates interoperable, state-of-the-art marking and detection across providers.
Track all of these on the live deadline tracker.
Penalties
Non-compliance with transparency obligations carries administrative fines of up to €15 million or 3% of total worldwide annual turnover, whichever is higher. Member-state market surveillance authorities enforce, and the first wave of enforcement attention is expected to focus on unlabeled deepfakes and unmarked synthetic media.
What to actually do
- Inventory every AI system you provide or deploy that touches EU users.
- Map each to the four obligations above (the scope checker does this).
- For generative systems, pick a marking approach — compare watermarking & provenance tools, or read the machine-readable marking guide.
- For chatbots, ship a clear disclosure in the interface.
- For published AI content, add visible labels and consider detection tooling for user uploads.
- Document all of it — governance platforms automate the evidence trail.
This guide is informational, maintained continuously as Commission guidance evolves, and is not legal advice. Last substantive review: August 2026.