Chapter V of the EU AI Act (Articles 51-56) introduces a dedicated regulatory framework for general-purpose AI (GPAI) models. This framework recognises that GPAI models present unique regulatory challenges because they are not designed for a single specific application but can be integrated into a wide variety of downstream AI systems and use cases.
Defining General-Purpose AI Models
Article 3(63) defines a "general-purpose AI model" as an AI model, including where trained with a large amount of data using self-supervision at scale, that displays significant generality and is capable of competently performing a wide range of distinct tasks. The key criteria are: (1) trained on large amounts of data; (2) significant generality; and (3) capability across a wide range of tasks. Most large language models (LLMs) and large multimodal models (LMMs) will qualify as GPAI models. The GPAI model category is separate from the AI system category -- a GPAI model may be integrated into an AI system by a downstream provider.
Regulation (EU) 2024/1689 — Article 51(1)
“In addition to the obligations listed in this Chapter, a general-purpose AI model shall be considered to present systemic risk if it: (a) has high impact capabilities evaluated on the basis of appropriate technical tools and methodologies, including indicators and benchmarks; or (b) based on a decision by the Commission, of its own initiative or following a qualified alert by the scientific panel.”
- GPAI models are regulated at the model level, not at the application/system level
- A GPAI model becomes a GPAI system when integrated with additional components for a specific application
- The provider of the GPAI model and the provider of the downstream AI system may be different entities
- Article 55 establishes a two-tier obligation structure: baseline obligations for all GPAI models + enhanced obligations for systemic risk models
- The GPAI provisions apply from 2 August 2025 (12 months after entry into force)