What each date actually is
| Date | What is due | Who it covers |
|---|---|---|
| 2 August 2026 | Article 50 in force: chatbot notice (50(1)), deepfake and public-interest text labels (50(4)), emotion / biometric notice (50(3)), and 50(2) marks on new launches | Providers and deployers of in-scope systems. A US, Canadian, or UK HQ does not take you out if the output is used in the EU. |
| 2 December 2026 | Article 50(2) machine-readable marking and detection only | Providers of generative systems placed on the market before 2 August 2026. Later launches do not get this date. |
| 2 December 2027 | Annex III stand-alone high-risk rules | Employment, education, credit, essential services, certain biometrics — not a support chatbot. |
The AI Act Service Desk says the same thing: transparency is enforceable from 2 August 2026, “although providers of AI systems placed on the market before that date have until 2 December 2026 to comply with marking and detection obligation under Article 50(2).”
If you ship a chatbot this week
- Put the notice in the product at the start of the first interaction, visible without a click. “This is an AI assistant. You are not speaking with a person.”
- Do not wait for 2 December. That date does not move Article 50(1).
- If the generative system was already on the market before 2 August, open a 2 December ticket for machine-readable marks. Decide this month: sign the Code of Practice, or document an equivalent method.
- If you launched on or after 2 August, the 50(2) mark is already due.
- Content generated before 2 August 2026 does not have to be labelled retroactively. The Commission still encourages it where possible.
What 50(2) is (and is not)
Article 50(2) is a provider duty: mark synthetic audio, image, video, or text in a machine-readable format so it can be detected as AI-generated or manipulated. It is not the human-readable chatbot notice, and it is not the deepfake label a person can see or hear. Deployers cannot rely on a hidden vendor watermark to satisfy Article 50(4).
The Commission FAQ carves out short codes, source code, machine-to-machine output that never reaches a person, closed-loop industrial work that is not the final output, and assistive standard editing that does not substantially change the input.
Signing the Code of Practice on Transparency of AI-generated Content is voluntary. The Commission has assessed it as adequate for Articles 50(2), (4), and (5). It does not operationalise the Article 50(1) chatbot notice.
Fines
The FAQ says fines can reach €15 million or 3% of total worldwide turnover for the preceding financial year, with proportionality for SMEs and small mid-caps. National market surveillance authorities handle most Article 50 cases. As of 27 August 2026 there was no official announcement of a first Article 50 fine or a named investigation against a specific startup.
Buy the checklist
The $29 Article 50 Ship Checklist is a one-page operator template: inventory, the exact notice line, deepfake labels, the 2 December marking ticket, and when to skip high-risk work. Print it and tick it.
Sources
Fetched 28 August 2026 (PT). No invented dates or quotes.
- European Commission, Article 50 FAQ, last update 24 July 2026. Transparency obligations under Article 50 of the AI Act
- AI Act Service Desk, “When does enforcement start?” ai-act-service-desk.ec.europa.eu
- Operator template built from the same pages: Article 50 Ship Checklist and Does Article 50 apply to my chatbot?