EU AI Act Deadlines: The 4 Dates That Decide What You Owe in 2026-2027

The EU AI Act phases in on four dates. Bans on the riskiest uses started February 2, 2025. Rules for general-purpose AI models started August 2, 2025. The big one is August 2, 2026, when transparency labels and most high-risk rules were set to apply. Product-embedded high-risk rules follow on August 2, 2027. A 2026 update called the Digital Omnibus is now moving some of the high-risk dates later.

What are the four EU AI Act deadlines?

The Act entered into force on August 1, 2024, then switches on in stages. These four dates are the schedule written into the law itself (Regulation 2024/1689, published in the Official Journal on July 12, 2024). This is law in force, not a proposal.

• February 2, 2025. Banned uses stop (Article 5). This covers social scoring, most workplace and school emotion recognition, and certain biometric tracking. AI literacy duties also begin (Article 4).
• August 2, 2025. Rules for general-purpose AI models apply (the large models behind chatbots and copilots). The EU governance bodies, national authorities, and penalty frameworks come online.
• August 2, 2026. Transparency rules (Article 50) apply. You must tell people when they are talking to AI and label AI-generated or manipulated content. High-risk rules under Annex III were also set for this date.
• August 2, 2027. Rules for high-risk AI built into regulated products, like machinery, lifts, and toys, take effect.

Sources: EU AI Act Service Desk timeline | Regulation 2024/1689 (Official Journal)

Did the August 2, 2026 deadline actually move?

Part of it. In November 2025 the European Commission tabled a package called the Digital Omnibus. Negotiators reached a provisional political agreement in May 2026. As a proposal, it is not final until it is formally adopted and published in the Official Journal, which was expected before August 2, 2026. Until that publication, the original dates are the law. Here is what the Omnibus changes if it lands:

• Transparency stays. The Article 50 disclosure rules still apply on August 2, 2026. This is the piece most consumer AI products will actually feel first.
• Standalone high-risk moves to December 2, 2027. The Annex III list (hiring, credit, biometrics, critical infrastructure, education, and more) gets a later start.
• Product-embedded high-risk moves to August 2, 2028. The August 2, 2027 date for AI inside regulated products slides a year.
• Sandbox deadline moves to August 2, 2027. Member states get more time to stand up regulatory sandboxes.

So plan for the transparency date as fixed, and treat the high-risk dates as later but not zero. The bans from 2025 and the general-purpose AI rules from 2025 are not touched.

Sources: Gibson Dunn, Omnibus agreement summary (May 2026) | European Parliament, MEPs support postponement (March 2026)

Is a US builder even in scope?

Often, yes. The Act reaches past EU borders. Where your company sits does not decide it. What decides it is a two-question in-scope test from Article 2.

1. Do you place the AI system on the EU market or make it available to users in the EU? If yes, you are in scope, even from a US address.
2. Is the output of your AI system used in the EU? If a provider or deployer sits outside the EU but the result is used inside the EU, that pulls you in too.

Selling a SaaS tool to a French customer, or running a model whose answers reach EU users, can be enough. A US LLC or Delaware C-corp does not opt you out. Where you choose to set up your company is a separate question from your EU exposure, and it does not reduce it.

Sources: AI Act, Article 2 (scope) | European Commission, regulatory framework for AI

What counts as high-risk, and does my product hit it?

High-risk is a defined list in Annex III, not a vibe. Your product is more likely to land there if it makes or shapes a decision about a person in one of these areas:

• Hiring, screening resumes, or evaluating workers.
• Credit scoring or deciding access to essential services.
• Biometric identification or categorization.
• Education and exam scoring.
• Critical infrastructure, migration, asylum, and border control.

A plain content generator or internal coding assistant usually is not high-risk, though it can still trigger the Article 50 transparency label. Hiring tools are the classic trap, and they draw scrutiny in the US too. If you build recruiting AI, treat the EU high-risk list and any US state hiring-audit rules as separate obligations you may owe at the same time.

Sources: Regulation 2024/1689, Annex III

What is my pre-August-2-2026 checklist?

Short version: confirm scope, ship the transparency basics, and track the high-risk clock. Do this before August 2, 2026.

1. Run the two-question scope test. Write down whether you place systems on the EU market or your output is used in the EU. Keep the answer on file.
2. Add AI disclosure. Tell users when they are interacting with AI. Label AI-generated or edited images, audio, video, and text. This is the August 2, 2026 Article 50 duty.
3. Check your model supplier. If you build on a general-purpose model, confirm the provider meets its August 2, 2025 obligations, since some duties flow to you.
4. Screen for high-risk use. Map your product against the Annex III areas above. If you touch hiring, credit, or biometrics, start documentation now even with the later date.
5. Watch the Official Journal. The Digital Omnibus deferrals are only real once published. Do not lock a plan on a proposal.

Sources: Regulation 2024/1689 | European Commission, regulatory framework for AI

If you want a second set of eyes on whether your build is in scope, that is the kind of thing I sort out at nomadtechnologist.com.

---

Not legal, financial, or tax advice.