Correction: this post has been updated. See what actually changed and what didn't for the current EU AI Act timeline.
Product teams have spent much of the year asking the same question. "Are we still doing this in August? I keep reading it might slip." The Digital Omnibus the Commission floated in November 2025 has since been finalised. The European Parliament and Council signed off on the AI Act simplification package, and the high-risk deadlines have officially moved. That does not mean August 2026 stopped mattering. It means it matters for a narrower reason than this post originally argued.
For compliance leads, that means re-pointing the plan, not abandoning it. The high-risk work still needs doing, just against the new dates below.
What actually switches on in August
Two things are still due on 2 August 2026, and a third has moved.
The first is enforcement against general-purpose AI model providers. Obligations under Chapter V have technically applied since August 2025, but the Commission's supervision and fining powers activate this August. Penalties top out at €15 million or 3 percent of global turnover, whichever is greater.
The second is Article 50 transparency. Chatbot disclosure, synthetic content marking, and deepfake labelling are all still due on 2 August 2026, unchanged by the Omnibus. If your product does any of those things, that date is the one to plan against.
What has moved is the full obligations regime for high-risk AI systems under Articles 9 to 17. Annex III stand-alone high-risk systems, covering recruitment, credit scoring, law enforcement, education and border control, now have until 2 December 2027. Annex I product-embedded high-risk systems, think medical devices, machinery and vehicles, move to 2 August 2028. Public sector deployers of high-risk systems get the same extended runway.
The Code of Practice is the path of least pain
The General-Purpose AI Code of Practice, finalised in 2025, is voluntary. Signing it does not exempt you from the Act. What it does is give you a structured way to demonstrate good-faith compliance, and the AI Office has signalled that signatories get a more collaborative posture during the first year of enforcement.
For teams shipping GPAI features, the Code is the cheapest lever you have. It maps obligations to evidence in a way the regulator already accepts.
What to do in the next 90 days
Don't stand down the high-risk work just because the deadline moved. Inventory which of your systems are high-risk under Annex III or Annex I, who in your supply chain is the provider versus the deployer, and where your evaluation evidence actually lives. December 2027 and August 2028 sound distant, but conformity assessment, technical documentation, and data governance take longer to build than most teams expect. Separately, if your product touches chatbots, synthetic content, or deepfakes, make sure Article 50 labelling is ready now. That one hasn't moved.
Want a faster way to evidence AI Act readiness across testing, evaluation, and post-market monitoring? EvalSpec helps teams map evaluations to specific AI Act articles and keep the audit trail in one place.
Try EvalSpec free - evaluate your AI system across all 5 dimensions. No code required.
Start your free evaluation