What it says
- Providers must design systems so that people know they are interacting with an AI from the first interaction, unless it is obvious. Systems that only work in the background or between machines are excluded.
- They must also add machine-readable marks that make it possible to detect content generated or manipulated with AI.
- Those who deploy the system must disclose deepfakes, AI-generated texts on matters of public interest, and the use of emotion recognition or biometric categorization.
- Published text that has gone through human review or editorial control does not need a label.
- Systems placed on the market before August 2, 2026 have until December 2, 2026 to comply with the marking obligation. Content generated before August 2 does not have to be labeled retroactively.
How it has evolved
Thread: Application of the AI Act. How the European AI Act is applied in practice: timeline, amendments, Commission guidelines, Spanish adaptation, and providers' response. Upcoming dates: December 2, 2026 (Article 50.2 marking for systems already on the market) and December 2, 2027 (Annex III high risk).
- June 8, 2026 · The Bureau of the Congress of Deputies commissions the committee report on the Organic Bill for the proper use and governance of AI (121/000096), with an amendment deadline of June 30.
- July 20, 2026 · AI Act transparency guidelines: what must be disclosed and labeled from August (this story)
- July 27, 2026 · The AI omnibus enters into force: high-risk rules are delayed to December 2027
- August 2, 2026 · The AI Act becomes generally applicable; the AI Office can exercise its enforcement powers over general-purpose models.
- October 5, 2026 · OpenAI will put an invisible watermark on ChatGPT texts in the EU
Commission questions and answers on Article 50: digital-strategy.ec.europa.eu
The Tech Guardian summarizes and comments in its own words on third-party publications. The full information belongs to the original source, which we recommend you read.