The EU AI Act's transparency rules just went live
EU AI Act transparency duties took effect Aug 2, with enforcement bodies now active
The three lines
- Transparency rules took effect August 2 — AI-generated content must be disclosed
- The EU AI Office and member-state authorities now hold live enforcement powers
- High-risk system duties were pushed to December 2027 and August 2028
Key questions
- What is the EU AI Act
- The world's first comprehensive AI law. It sorts AI by risk — banned, high-risk, transparency-required, and unrestricted — and has been phasing in since 2024. This week the transparency tier and the enforcement machinery went live.
- Does it apply to non-EU companies
- Often, yes. The Act reaches any provider whose AI serves EU users, regardless of where the company sits — extraterritoriality is its signature feature. Practical guidance on exactly which services fall where is still being issued by the Commission.
- What happens to violators
- Fines scaled to global revenue are on the books. But enforcement only just switched on — there are no precedent cases yet, so the real-world strictness of the regime is the thing to watch, not assume.
The world's biggest AI regulation went operational this week. The EU AI Act's transparency rules took effect on August 2 — and on the same day, the EU AI Office and the authorities of 27 member states formally acquired supervision and enforcement powers. The rehearsal period is over: violations can now become investigations.
1. What changed
Two things. First, transparency duties: content generated by AI must be identifiable as such, and AI systems that converse with humans must disclose that they are AI. The European Commission has been issuing practical guidance on how AI-generated content should be marked.
Second, enforcement went live. Until now the Act had rules but no active referee; as of August 2, the AI Office and national authorities are responsible for implementing, supervising and enforcing it.
2. The timeline — where things stand
| Date | Milestone | Status |
|---|---|---|
| 2025 | bans and baseline duties | in force |
| Aug 2, 2026 | transparency rules + enforcement live | this week |
| Dec 2, 2027 | high-risk (Annex III) duties | extended deadline |
| Aug 2, 2028 | high-risk (Annex I) duties | extended deadline |
| ongoing | AI Omnibus adjustments, guidance | fluid |
The high-risk postponements are the other half of the story: the EU Council approved timetable changes under the AI Omnibus — switching the regime on while giving industry longer to comply with its heaviest tier.
3. What is still open
For companies outside Europe, the practical question is reach: the Act's extraterritorial design means a service in Seoul or San Francisco is in scope the moment EU users are. But until first enforcement cases land, the regime's actual teeth are unmeasured. How regulation interacts with the breakneck model-release race is the subject of today's companion piece.