Determine whether a model you train or fine-tune is a general-purpose AI model under the EU AI Act and whether you are its provider

Documented steps

1. Read the definition of 'general-purpose AI model' in Article 3(63) of Regulation (EU) 2024/1689: a model displaying significant generality, competently performing a wide range of distinct tasks, and capable of being integrated into downstream systems. Text: https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-3
2. Read the definition of 'provider' in Article 3(3): the person that develops the model (or has it developed) and places it on the market or puts it into service under its own name or trademark, whether for payment or free of charge.
3. Open the Commission's official 'Guidelines for providers of general-purpose AI models' (adopted July 2025) and work through its scope and provider-determination sections: https://digital-strategy.ec.europa.eu/en/policies/guidelines-gpai-providers . Download the full PDF from that page — the landing page alone does not contain the indicative compute and modification criteria.
4. If you are fine-tuning or otherwise modifying a third-party GPAI model, apply the Guidelines' criteria for when a downstream modifier becomes the provider of a new model. Read the exact indicative compute criterion in the Guidelines PDF and in Recital 109 of the AI Act; do not rely on figures quoted in secondary commentary, which vary.
5. If you become the provider of a GPAI model, confirm Chapter V obligations (Articles 53-55) apply. These have applied since 2 August 2025 under Article 113 and were NOT postponed by the AI Omnibus.
6. Check the open-source carve-out in Article 53(2): releasing under a free and open-source licence with publicly available parameters, weights and architecture information exempts you from Article 53(1)(a) and (b) documentation duties only. Article 53(1)(c) copyright policy and 53(1)(d) training-content summary still apply, and the carve-out does not apply at all to models classified as having systemic risk.
7. Separately test whether the model crosses the Article 51 systemic-risk threshold, which triggers Article 55 obligations on top of Article 53.
8. Record your determination — model classification, provider status, applicable articles and the reasoning — because the AI Office and market surveillance authorities can request it under the Article 91 information-request power.
9. Re-run the determination on any material retraining, compute scale-up, licence change or rebranding, since provider status and systemic-risk status can both change over a model's lifecycle.