The EU AI Act on 2 August 2026 — what applies now
If you saved an EU AI Act timeline in 2024 or 2025, it is now wrong. In June 2026 — weeks before the Act’s biggest application date — the EU adopted the “Digital Omnibus on AI”, a package of amendments to Regulation (EU) 2024/1689. The European Parliament approved it on 16 June 2026, the Council on 29 June 2026, with publication in the Official Journal expected before the end of July. Here is what 2 August 2026 still activates, and what moved.
The short version
On 2 August 2026, the AI Act’s transparency rules start to apply, and national authorities begin supervising the AI literacy obligation. The heaviest part of the Act — the obligations for high-risk AI systems — moved: 2 December 2027 for standalone high-risk systems, 2 August 2028 for AI embedded in regulated products.
What still starts on 2 August 2026
Transparency (Article 50). People must be told when they interact with an AI system, unless that is obvious from context. Synthetic audio, images, video, and deep fakes must be disclosed. If your product includes an assistant that talks to customers, or your marketing uses synthetic media, this is your date. One narrow grace period: systems already on the market before 2 August 2026 have until 2 December 2026 to implement machine-readable marking of synthetic content.
Supervision of AI literacy (Article 4). The literacy obligation has applied since 2 February 2025, but the European Commission’s AI literacy Q&A confirms that supervision and enforcement by national market surveillance authorities start on 2 August 2026. There is no dedicated fine for Article 4 — enforcement runs through national rules, and sanctions must be proportionate. Note that Article 4 itself was rewritten by the Omnibus: the duty is now to support AI literacy, not to ensure a level of it.
Already in force since earlier dates: the prohibitions (including emotion recognition at the workplace) since 2 February 2025, and the rules for general-purpose AI models plus the penalties framework since 2 August 2025. Those dates did not move.
What moved, and where
- Standalone high-risk AI systems (Annex III) — recruitment and HR tools, credit scoring, education, access to essential services — moved from 2 August 2026 to 2 December 2027.
- AI embedded in regulated products (Annex I) — machinery, medical devices, toys — moved to 2 August 2028.
- National AI regulatory sandboxes now need to be operational by 2 August 2027 instead of 2026.
- The Omnibus also tightened one thing: a new prohibition on AI systems built in a way that generating non-consensual intimate imagery or child sexual abuse material is a foreseeable outcome. It applies from 2 December 2026.
Analyses by Gibson Dunn and Mishcon de Reya walk through the full amendment set.
What this means for a normal company
If you deploy AI rather than build it, the practical readings are:
- Your customer-facing assistant needs a disclosure check now. “Talking to an AI” must be evident. This is cheap to fix and visible to any regulator or competitor who looks.
- Your HR stack got time, not absolution. High-risk duties for hiring and performance tools now bite in December 2027 — but the prohibition on emotion recognition at work has applied since February 2025. Don’t confuse the two clocks.
- Literacy supervision starts even though the duty softened. The question an authority can ask from August is not “what score did your staff reach” but “what measures did you take” — which makes your records the whole answer.
Do this quarter
- Inventory which AI systems are actually in use, approved or not, and what data flows into them.
- Check disclosures on anything AI-driven that your customers and employees interact with.
- Run a literacy program that leaves evidence — role-based measures, recorded per person. If a vendor is still selling you urgency with claims that stopped being true in June, read the amended text first.
Our AI Act readiness program turns your policy and tools into role-based learning with evidence you can export when someone asks. If you want to map your own rollout against the new dates, book a walkthrough and bring your AI inventory.
Ulern builds readiness and evidence. This post explains the timeline in plain terms — it is not legal advice.