The GPAI regime: what providers of general-purpose AI models must already do
Since 2 August 2025, providers of general-purpose AI models face their own rules, with extra requirements for systemic risk. With the July 2025 Code of Practice, the mandatory training-data template and the AI Office's enforcement powers from 2 August 2026, the regime is complete.
Download the regime cheat sheet (PDF) ↓
While the high-risk obligations are expected to shift to late 2027, the regime for general-purpose AI models (GPAI) is already running at full speed. Chapter V has applied to new models since 2 August 2025, and on 2 August 2026 the European AI Office gains the power to enforce it. For anyone offering large models โ or building on them โ this is currently the most concrete part of the AI Act.
The baseline obligations (Article 53)
Every provider of a GPAI model must: draw up and maintain technical documentation; provide information and documentation to parties integrating the model into their own AI systems; operate a policy to comply with European copyright law, including respecting text-and-data-mining reservations; and publish a public summary of the training content. For that summary a mandatory template has been available since 24 July 2025, asking among other things for the main data sources and domain names.
Additional requirements for systemic risk (Article 55)
Models with systemic risk โ presumed above a training compute of 10^25 FLOPs, or upon designation by the Commission โ carry a second layer: model evaluations including adversarial testing, assessment and mitigation of systemic risks at Union level, reporting of serious incidents to the AI Office, and an adequate level of cybersecurity for the model and its infrastructure.
The Code of Practice as the compliance route
The GPAI Code of Practice of 10 July 2025 translates these open obligations into concrete measures in three chapters: transparency, copyright and safety. Signing is voluntary, but signatories get a simplified supervisory path โ adherence to the Code counts as the accepted route to Chapter V compliance. Major providers, including Microsoft and Google, have signed; the signatories have formed a Signatory Taskforce chaired by the AI Office to apply the Code consistently.
Enforcement and the date that matters
From 2 August 2026 the AI Office can request information, order evaluations, impose measures and issue fines of up to โฌ15 million or 3 percent of worldwide annual turnover (Article 101). Models already on the market before 2 August 2025 must comply in full by 2 August 2027. For European organisations that build no models of their own, the regime is indirect but essential: their suppliers' documentation and training-data summaries will be the basis on which they can demonstrate their own diligence โ from procurement to literacy.
Sources
- https://eur-lex.europa.eu/eli/reg/2024/1689/oj
Regulation (EU) 2024/1689, Chapter V (Articles 51-56) and Article 101 (fines for GPAI providers). - https://digital-strategy.ec.europa.eu/en/policies/contents-code-gpai
The GPAI Code of Practice, published 10 July 2025, with chapters on transparency, copyright and safety. - https://digital-strategy.ec.europa.eu/en/news/commission-presents-template-general-purpose-ai-model-providers-summarise-data-used-train-their
The mandatory template for the public summary of training content, presented 24 July 2025. - https://digital-strategy.ec.europa.eu/en/policies/signatory-taskforce-gpai-code-practice
The Signatory Taskforce of Code signatories, chaired by the AI Office.
Read next
The AI Office: role, tasks and enforcement powers
The AI Office within the European Commission coordinates implementation of the AI Act and is the exclusive supervisor of GPAI models. It draws up codes of practice, conducts investigations and can have fines imposed on model providers.
GPAI enforcement goes live on 2 August 2026 โ and the Signatory Taskforce
From 2 August 2026 the European Commission can enforce the GPAI model rules, fines included (Art. 101). Obligations have applied since 2 August 2025; older models comply by 2 August 2027. The Signatory Taskforce, chaired by the AI Office, steers the Code of Practice.
The AI Act's expert bodies: the Scientific Panel and the Advisory Forum
On 1 June 2026 the European Commission appointed the two expert bodies the AI Act foresees: a 60-member Scientific Panel of independent experts (Art. 68) advising on general-purpose AI and systemic risk, and a 174-member Advisory Forum (Art. 67) for broad input. Both serve two-year terms.
