The EU AI Act starts fining GPAI model providers on 2 August 2026
The high-risk deadline moved to 2027, but 2 August 2026 still holds for one group: general-purpose AI model providers. From that day the Commission can enforce their rules and fine them up to 3 percent of worldwide turnover or 15 million euros.
A lot of AI Act timelines got a year longer this spring, so it is easy to assume 2 August 2026 stopped mattering. For most obligations the high-risk ones, it did move. For one group it did not: providers of general-purpose AI (GPAI) models. From 2 August 2026 the European Commission can enforce their rules and fine them.
What switches on that day
GPAI model obligations have technically applied since 2 August 2025, but for the first year there was no one enforcing them. That grace year ends on 2 August 2026, when the Commission, acting through the AI Office, gains its supervision and enforcement powers over GPAI providers. From that date it can request documentation and information, evaluate a model, require compliance and risk-mitigation measures up to withdrawing a model from the market, and impose fines.
The GPAI fine sits on its own track
GPAI model providers are not fined under the general Article 99 tiers. Article 101 gives them a separate ceiling: up to 15 million euros or 3 percent of worldwide annual turnover, whichever is higher. That is the same percentage as the general non-compliance tier, but it is a distinct provision aimed squarely at the model makers, and it is separate again from the 7 percent or 35 million euro ceiling that applies to prohibited AI practices.
Who this actually hits, and who it does not
This is aimed at the companies that build and place general-purpose AI models on the market, the foundation-model providers, not most businesses that simply use AI inside a product. If you fine-tune or deploy someone else’s model, your first-order duty is the Article 50 transparency rules, which also start on 2 August 2026, rather than the GPAI provider obligations. One date worth noting either way: GPAI models already on the market before 2 August 2025 have until 2 August 2027 to be brought into line.
Why this date held when the high-risk one moved
The digital omnibus, agreed politically on 7 May 2026 and since adopted by the European Parliament on 16 June 2026 and the Council on 29 June 2026 (with publication in the Official Journal still to come), pushed the high-risk deadlines back: stand-alone Annex III systems to 2 December 2027 and AI inside regulated products under Annex I to 2 August 2028. It left 2 August 2026 in place as the day GPAI enforcement powers, the Article 50 transparency rules, and national market-surveillance powers all switch on. The AI Office has also said it will treat signatories of its GPAI Code of Practice who act in good faith as compliant while they finish implementing, rather than penalising them on day one.
Whether any of this reaches your turnover is a question of which violation, if any, applies to you. Our free fine calculator models all three Article 99 tiers and the separate Article 101 GPAI track, and applies the lower-of cap for SMEs, so you can enter your worldwide turnover and see which leg would actually bind.
Put it to work on your own case
The free tool below turns this into a result for your situation, in your browser, with no signup.
Estimate an AI Act fineSources
More updates
The EU AI Act transparency rules take effect on 2 August 2026
From 2 August 2026, Article 50 of the EU AI Act requires you to tell people when they are dealing with an AI system and to label AI-generated content. The high-risk deadlines were pushed back. This one was not.
The EU AI Act high-risk deadline is no longer 2 August 2026
A May 2026 political agreement pushed the EU AI Act high-risk obligations back by more than a year. Stand-alone high-risk systems now have until 2 December 2027, and AI inside regulated products until 2 August 2028.
The EU AI Act just gained a new banned practice, and the omnibus that added it is now adopted
The digital omnibus that reshaped the AI Act cleared its final legislative step when the Council adopted it on 29 June 2026. Beyond the delayed high-risk deadlines, it adds a new prohibited practice under Article 5: AI built to generate non-consensual intimate imagery, with a 2 December 2026 compliance date.