Since 2 August 2026, the European Commission's enforcement powers over general-purpose AI providers have been in application. The Commission says it will enforce providers' obligations "including with fines." People often misread that date. The EU AI Act's GPAI obligations are not new duties. They are the duties in Article 53, and the Transparency and Copyright chapters of the GPAI Code of Practice offer providers a way to demonstrate compliance with those Article 53 obligations, a code that was published on July 10, 2025. What changed in August is that someone can now enforce them.
Three dates, one dividing line
The Commission's timeline is short. From 2 August 2026, the Commission's enforcement powers apply. By 2 August 2027, providers of GPAI models placed on the market before 2 August 2025 must comply. That makes 2 August 2025 the date the core GPAI obligations began applying, with providers of models placed on the market before that date given until 2 August 2027 to comply.
The Commission's guidelines page puts it plainly: the obligations for providers of GPAI models entered into application on 2 August 2025. Enforcement of them waited a year.
The structure is worth reading closely. The extra year in the Commission's timeline applies only to models that were on sale before the 2025 cutoff. The timeline gives no comparable grace period to models launched after that date. That is an inference from the schedule, but it is hard to read it any other way.
Before enforcement began, the Commission's role was mainly supportive. The AI Office supports providers in their compliance, especially the Code of Practice signatories. That support continues, but the Commission can now also enforce the rules, including with fines.
| Date | Event | Impact |
|---|---|---|
| 10 July 2025 | GPAI Code of Practice published | Voluntary route to compliance |
| 2 August 2025 | Market cutoff for "legacy" models | Older models get extra time |
| 2 August 2026 | Commission enforcement powers apply | Fines become possible |
| 2 August 2027 | Deadline for pre-cutoff models | Legacy models must comply |
Article 53's four duties, read as a buyer
Article 53(1) sets four duties for providers:
- Technical documentation. Providers must draw up and keep up-to-date technical documentation of the model, including its training and testing process and the results of its evaluation. It must cover at least what Annex XI lists and go to the AI Office on request.
- Information for downstream providers. Providers must make information and documentation available to providers of AI systems who intend to integrate the model. It has to be good enough to give those integrators a clear understanding of the model's capabilities and limitations, with Annex XII as the minimum.
- Copyright policy. Providers need a policy to comply with Union law on copyright. That includes using state-of-the-art technologies to identify and respect rights reservations under Article 4(3) of Directive (EU) 2019/790.
- Training-data summary. Providers must make publicly available a sufficiently detailed summary about the content used for training.
| Article 53(1) duty | Audience | Open-source exempt? | Code chapter |
|---|---|---|---|
| (a) Technical documentation | AI Office, authorities | Yes, unless systemic risk | Transparency |
| (b) Downstream information | Integrating providers | Yes, unless systemic risk | Transparency |
| (c) Copyright policy | Rights holders | No | Copyright |
| (d) Training-content summary | The public | No | — |
For product leads, duty (b) gets less attention than it deserves, and it is written for them.
The open-source exemption is narrower than it looks
The obligations set out in paragraph 1, points (a) and (b), shall not apply to providers of AI models that are released under a free and open-source licence that allows for the access, usage, modification, and distribution of the model, and whose parameters, including the weights, the information on the model architecture, and the information on model usage, are made publicly available. The exemption covers only points (a) and (b), the two documentation duties. Open-weight providers still need a copyright policy and a public training-content summary. The exception also shall not apply to general-purpose AI models with systemic risks.
Assuming "open" means "exempt" is an easy mistake to make. Teams fine-tuning open models get some relief from the Commission's guidelines, which say only those making significant modifications to AI models need to comply with the obligations for providers of GPAI models, not those making minor changes. Where that line falls is set out in the guidelines, and teams should check it there rather than assume.
Three chapters, two kinds of provider
The Code of Practice has three separately authored chapters: Transparency, Copyright, and Safety and Security. The first two offer all providers of general-purpose AI models a way to demonstrate compliance with their obligations under Article 53. The Transparency chapter comes with a Model Documentation Form that turns the documentation duty into a fill-in template. The Safety and Security chapter sets out state-of-the-art practices for managing systemic risks, i.e. risks from the most advanced models, and only those providers need it.
The Commission and the AI Board have confirmed the code is an adequate voluntary tool for showing compliance. According to the Commission, signing reduces their administrative burden and gives them more legal certainty than other ways of proving compliance. The cost of not signing backs this up. Providers that don't follow an approved code or meet a harmonised standard shall demonstrate alternative adequate means of compliance for assessment by the Commission. They must also file reports on how they intend to comply through the EU SEND platform. Harmonised standards, by contrast, give a presumption of conformity for the obligations they cover.
The Commission's code page lists the signatories: AI Studio Delta, Aleph Alpha, Almawave, Amazon, Anthropic, Black Forest Labs, Bria AI, Cohere, Domyn, Dweve, Fastweb, Google, IBM, LINAGORA, Microsoft, Mistral AI, Open Hippo, OpenAI, Pleias, Reflection AI, Inc., ServiceNow and WRITER. It adds that xAI signed up to the Safety and Security Chapter only, so xAI has to show compliance on transparency and copyright by other adequate means. Signatories also formed a Signatory Taskforce chaired by the AI Office; its Vademecum holds the rules of procedure and the list of members.
What to ask your provider
None of this is legal advice. But since August, the documents Article 53 requires are ones a regulator can actually enforce. In practice, that means four questions for your provider:
- Are they on the Commission's signatory list?
- Will they share Annex XII-level integration documentation?
- Is there a public training-content summary?
- For open-weight models, where is the copyright policy?
The Commission's guidelines on the scope of obligations are available in all 24 official EU languages, so language is no reason to skip them. A provider that can't answer these questions now has a regulator who can ask them the same things.
Sources
- digital-strategy.ec.europa.eu – "Guidelines for providers of general-purpose AI models"
- digital-strategy.ec.europa.eu – "The General-Purpose AI Code of Practice"
- artificialintelligenceact.eu – "Article 53: Obligations for Providers of General-Purpose AI Models | EU Artificial Intelligence Act"