The Guidelines were developed to clarify the application of Chapter V of the AI Act (Regulation (EU) 2024/1689), providing guidance on aspects essential for the implementation and understanding of the AI Act relating to general-purpose AI models (GPAI). Although the Guidelines have been published, they will be formally adopted by the European Commission at a later date.
Of particular note, the Guidelines provide important clarifications on the following points:
- Criteria for placing a GPAI model on the market: The Guidelines set out several scenarios that constitute the placing of a general-purpose AI model on the Union market, including, for example, making a model available through an API, uploading it to a public repository, or using a GPAI for certain internal processes that affect third parties or the rights of individuals within the Union. In this context, the Guidelines reference both the relevant definitions from the AI Act and the ‘Blue Guide’ on the implementation of EU product rules (2022/C 247/01).
- Modification and new provider status: Consistent with the Blue Guide, the Guidelines also clarify the threshold at which a downstream actor modifying a GPAI model is considered a new provider. While not explicitly stated, the approach appears to take inspiration from Article 25 of the AI Act on substantial modification. In the context of AI systems, substantial modification can trigger provider obligations for the AI operator. For AI models, the Guidelines specify that a downstream modifier becomes the provider of the modified general-purpose AI model only if the modification leads to a significant change in the model’s generality, capabilities, or systemic risk.
- Systemic risk presumption: Notably, the Guidelines confirm that if a general-purpose AI model with systemic risk is modified in such a way that the downstream actor becomes the provider of the modified model, the resulting model is presumed to possess high-impact capabilities and is therefore also classified as a general-purpose AI model with systemic risk in accordance with the AI Act.
- Scope of documentation and disclosure obligations for modifications: In line with Recital 109 of the AI Act, the Guidelines clarify that when a general-purpose AI model is modified or fine-tuned, the provider’s obligations are limited to that specific modification or fine-tuning. For instance, technical documentation required under Article 53(1), points (a) and (b), is limited to information on the modification itself, including any new training data sources. Similarly, the copyright policy under Article 53(1), point (c), and the summary of content used for training under Article 53(1), point (d), are limited to data used as part of the modification.
Finally, the Guidelines state that the Commission’s approach may evolve in response to technological, societal, or market developments. The full text of the Guidelines is available here.
