Every AI Policy Ever · open in the interactive register

Published (non-binding, current)Non-binding · Guidelines / framework

Guidelines on the scope of obligations for providers of general-purpose AI models under the AI Act

Record EU-11

Issued by
European Commission
Jurisdiction
European Union
Level · region
Supranational · Europe
Published / adopted
2025-07-18
Last updated
2026-04-20
Language
All 24 official EU languages

Summary

Commission guidelines setting out its interpretation of the AI Act's obligations for providers of general-purpose AI models, which apply from 2 August 2025. A model counts as general-purpose if it was trained with compute exceeding 10^23 floating point operations and can generate language, text-to-image or text-to-video. Only an actor making significant modifications to a model, not minor changes, becomes its provider; the guidelines also set out when providers of free and open-source models are exempt from certain obligations and what adhering to the General-Purpose AI Code of Practice means for enforcement. They are not legally binding but will guide enforcement.

Safety topics

general-purpose / frontier AIcatastrophic risk

Status notes

Published on 18 July 2025, ahead of the entry into application of the obligations for providers of general-purpose AI models on 2 August 2025. A Commission Communication, available only in English, approved the content of the guidelines, which are available in all 24 official EU languages. Providers of models with systemic risk, which include risks to fundamental rights and safety and risks related to loss of control over the model, must assess and mitigate these risks and notify the AI Office without delay. The Commission's enforcement powers, including fines, apply from 2 August 2026, and providers of models placed on the market before 2 August 2025 must comply by 2 August 2027. The guidelines were developed through a public consultation and reflect input from an expert pool set up by the Commission's Joint Research Centre. The Commission's page for the guidelines shows a last update of 20 April 2026.

Sources

Open the official source https://digital-strategy.ec.europa.eu/en/library/guidelines-scope-obligations-providers-general-purpose-ai-models-under-ai-act

Secondary source: https://digital-strategy.ec.europa.eu/en/policies/guidelines-gpai-providers

Local copy

Archived. A source changed since the previous snapshot; both versions are kept. Last attempt 2026-10-17 18:20:41 UTC

Official source (third-party copy, not published) archive/EU-11/2026-10-11/official.html
Retrieved 2026-10-11 13:58:34 UTC
SHA-256 daca7f4eeb4b7425ee618eb723a9dd990866f9912a565a37147970a64ab46517
Linked document (PDF) (third-party copy, not published) archive/EU-11/2026-10-17/document-1.pdf
Retrieved 2026-10-17 16:46:56 UTC
SHA-256 a57214ae2c6a671ebbe5d28d00eb90c0df8eb415e60f5a177d6bebb6f01603fe
Secondary source (third-party copy, not published) archive/EU-11/2026-10-11/secondary.html
Retrieved 2026-10-11 13:58:51 UTC
SHA-256 f5296983ac5ad517740fe8cb35333197ad4b9b05dd1ddc5e585a1987d2c0b95c

Verification

A
The instrument's own official text, or an official register, was stored and confirms the title, date and status. checked 2026-10-17 · official-url-fetched
Re-read in the recheck of 17 Oct 2026: the Commission library page (official, stored 11 Oct 2026) confirms the title, publication on 18 July 2025, the 24 language versions, the approving Communication and a last update of 20 April 2026; the Commission's overview page of the guidelines (stored 11 Oct 2026, last update 28 April 2026) confirms their key points, their non-binding character and the application and enforcement timeline; the Commission's explanatory item of 18 July 2025 (stored in this run) gives the indicative compute criterion, the four key topics, the systemic-risk obligations and the consultation. The guidelines' text and the approving Communication are offered only through ec.europa.eu/newsroom/dae/, which ec.europa.eu's robots.txt disallows, so neither is stored. Secondary sources give the Communication's number as C(2025) 5045 final; it is not shown in a stored official file, so official_identifier stays empty. Version 1's remark that the technical thresholds were not checked moved here from notes.
Second check: Independent check by Claude Code (claude-sonnet-5-5), 2026-10-17: 8 field(s) supported.

History

  1. 2026-10-17notes: Published 18 July 2025, ahead of the GPAI obligations applying on 2 Aug 2025. Th → Published on 18 July 2025, ahead of the entry into application of the obligation — rewritten from the stored files; systemic-risk duties, the consultation and the JRC expert pool added; the process remarks (C number not shown, thresholds not checked) moved to verification.note and open_questions2026-10-17-recheck-europe-1 · Claude Code (claude-opus-5-5)
  2. 2026-10-17safety_topics: ["general-purpose / frontier AI", "catastrophic risk", "evaluation & testing"] → ["general-purpose / frontier AI", "catastrophic risk"] — 'evaluation & testing' dropped: no stored text of this record shows evaluation or testing content; 'catastrophic risk' kept for models with systemic risk, which include risks of loss of control2026-10-17-recheck-europe-1 · Claude Code (claude-opus-5-5)
  3. 2026-10-17summary: Explains when a model counts as a general-purpose AI model and when it is a GPAI → Commission guidelines setting out its interpretation of the AI Act's obligations — rewritten from the stored library and overview pages and the Commission's explanatory item: the indicative 10^23 FLOP criterion, modifications, the open-source exemption and the link to the Code of Practice2026-10-17-recheck-europe-1 · Claude Code (claude-opus-5-5)

Related records

EU AI Act European Union · Partially in force

GPAI Code of Practice European Union · Published (non-binding, current)

GPAI training-data summary template European Union · Published (non-binding, current)

Citation

European Commission. Guidelines on the scope of obligations for providers of general-purpose AI models under the AI Act. 2025-07-18. https://digital-strategy.ec.europa.eu/en/library/guidelines-scope-obligations-providers-general-purpose-ai-models-under-ai-act. Status: Published (non-binding, current); verified 2026-10-17 (level A). Local copy archive/EU-11/2026-10-11/official.html retrieved 2026-10-11 13:58:34 UTC, SHA-256 daca7f4eeb4b7425ee618eb723a9dd990866f9912a565a37147970a64ab46517. Every AI Policy Ever, version 1.18 (snapshot 2026-10-18), record EU-11.

Improve this record

Know of an amendment, a status change or an error? Report news or an error on GitHub; the form opens with this record filled in.