The AI Act did not switch on all at once on August 2, 2026. That date is one milestone in a staggered calendar: the one at which enforcement begins for rules that already applied, such as the prohibitions from February 2025 and the obligations for general-purpose AI models from August 2025, and at which the transparency duties of Article 50 apply for the first time.
Who enforces what
The Commission’s AI Office enforces the rules for providers of general-purpose AI models, including those with systemic risk. National market-surveillance authorities handle the rest, and the European Data Protection Supervisor covers AI used by EU institutions. The Commission also opened a complaint tool, a whistleblower channel and a route for companies that build on another provider’s general-purpose model.
Four transparency cases
Article 50 is not a duty that falls on every AI product. The Commission structures it as four cases, two for providers and two for deployers:
- Providers of systems that talk to people must design them so that people know they are dealing with AI, unless that is obvious to a reasonably well-informed person.
- Providers of systems that generate audio, images, video or text must mark the output in a machine-readable format that can be detected as artificial. Tools that only assist standard editing and do not substantially alter the input are exempt.
- Deployers of emotion-recognition or biometric-categorisation systems must inform the people exposed to them.
- Deployers who publish deepfakes must disclose them as artificial, and those who publish AI-generated text to inform the public on matters of public interest must label it, unless it went through human review under editorial responsibility.
The Commission published guidelines on July 20 to define those notions, and Article 50 requires the information to reach people “in a clear and distinguishable manner” at the latest at their first interaction or exposure.
Deadlines and fines
Generative systems already on the market before August 2 have until December 2, 2026 to comply with the marking duty, according to the Commission’s Article 50 FAQ. Deepfakes made before August 2 need no retroactive label.
Breaching Article 50 can cost up to €15 million or 3% of worldwide annual turnover, whichever is higher, under Article 99; for small companies the lower of the two applies. Violations of the prohibitions sit higher, at €35 million or 7%.
The code of practice
A voluntary code of practice on AI-generated content, published on June 10, 2026 and assessed as adequate by the Commission and the AI Board, is a way to meet the marking and labelling duties. Its first section covers providers and its second, deployers. Companies that do not sign it must show compliance by other equivalent means.
What is still to come
The AI Omnibus, in force since July 27, moved the remaining milestones: high-risk rules under Annex III now apply from December 2, 2027, and high-risk AI inside regulated products from August 2, 2028. Obligations for general-purpose models under Articles 53 and 55 are a separate regime that applied from August 2025 and became enforceable on the same August 2.
Related reading
- Guidelines on transparency obligations for providers and deployers — European Commission
- Article 50 of the AI Act — AI Act Service Desk
- Fable 5.1 and Mythos 5.1 — IA al Día