Was the EU AI Act delayed? What actually applies from 2 August 2026

“I heard the AI Act got postponed” is the single most dangerous sentence in AI compliance right now. Parts of the timeline have been genuinely debated in Brussels — which is exactly why so many teams wrongly concluded the whole thing slipped. Here is what the law as in force actually says.

The staged timeline in Article 113

Regulation (EU) 2024/1689 entered into force on 1 August 2024 and applies in stages:

• 2 February 2025 — prohibitions (Article 5) and AI literacy obligations.
• 2 August 2025 — governance rules and obligations for general-purpose AI models.
• 2 August 2026 — the general application date, including the Article 50 transparency obligations for chatbots and synthetic content.
• 2 August 2027 — extended deadline for high-risk systems under Article 6(1) (products covered by EU harmonisation legislation).

What the “delay” talk is actually about

The European Commission's digital-omnibus simplification package proposed easing and re-sequencing parts of the AI Act — most notably around high-risk system obligations and their linkage to standards availability. Two things matter for your planning:

1. A Commission proposalis not law. Until the co-legislators adopt an amending act and it is published in the Official Journal, Article 113's dates stand.
2. The discussions centre on high-risk classification and documentation burdens — not on the basic transparency duties of Article 50. Telling users they are talking to an AI is precisely the kind of low-burden rule simplification packages leave in place.

Planning on a rumoured delay is a bet with asymmetric payoff: if you are wrong, you are non-compliant on day one of a regime with fines up to €15M or 3% of worldwide turnover for transparency breaches (Article 99(4)); if you are right, you spent a few hours adding disclosures users increasingly expect anyway.

What applies to a typical SaaS product on 2 August 2026

• Chatbot / voice-agent disclosure — Article 50(1). See is my chatbot covered?
• Machine-readable marking of AI-generated content — Article 50(2), if you provide generation features (see the C2PA guide).
• Deep-fake and public-interest text disclosure — Article 50(4), if you publish such content (see labelling AI-generated content).
• These apply to non-EU companies serving EU users too — details here.

How to use the time you have

Article 50 compliance is unusual among EU tech regulation in being genuinely fast to implement: an inventory, a few disclosure strings, and marking for generated media. Work through the step-by-step checklist, or start with the free 3-minute check to see which paragraphs apply to your product at all.