AI Act, Article 50 — transparency obligations
Providers and deployers of AI systems placed on the market or used in the EU. Article 50 splits by role: paragraphs 1 and 2 bind the PROVIDER, paragraphs 3 and 4 bind the DEPLOYER. No size threshold.
- 01As deployer of an emotion-recognition or biometric-categorisation system: inform the people exposed to it that it is operating, and process their data under the GDPR.
- 02As deployer publishing AI-generated or manipulated image, audio or video: disclose that the content is artificial.
- 03Provide that information clearly and at the latest at the moment of first interaction or exposure.
- 01As provider of a system that interacts directly with a person: design it so the person is informed they are interacting with an AI, unless that is obvious.
- 02As provider of a system generating synthetic audio, image, video or text: mark the output in a machine-readable format detectable as artificially generated.
- 03Generative systems already on the market before 2 August 2026 have until 2 December 2026 to comply with the Article 50(2) marking duty (new Art. 111(4)).
- 01No general 'we used AI to assess you' notice. Article 50(1) bites on systems a person interacts with — a recruiting chatbot or an AI interviewer — not on a CV-screening engine the candidate never converses with. That gap is exactly what Ontario s. 8.4 fills.
- 02No conformity assessment, no registration in the EU database, no risk-management system. Those are the Chapter III high-risk duties, deferred to 2 December 2027 (see EU-02).
- 03No bias audit and no human-review right.
- 04The text-disclosure duty in 50(4) applies only to text published to inform the public on matters of public interest, and not where a human exercised editorial control.
Set by Member States under Article 99. Breach of Article 50 transparency duties is subject to administrative fines up to EUR 15 000 000 or 3% of total worldwide annual turnover, whichever is higher, with proportionate caps for SMEs.
Two exposures, both ours rather than a customer's. First: our terms of service and privacy policy still describe voice-tone and emotional-tone analysis the product does not perform. If the documents were accurate, deployers of BitStage would be pulled into the Article 50(3) emotion-recognition notice duty that took effect on 2 August. They are not accurate, and they are still live — open since 22 July. Second: The Signal Minute is presented by a synthetic anchor. We disclose that in words on every episode, which is more than most, but Article 50(2) requires machine-readable marking, and the transitional in Article 111(4) gives us until 2 December 2026 to do it. We have not done it yet.
“the natural persons concerned are informed that they are interacting with an AI system”
Reg. (EU) 2024/1689, Art. 50(1)