New JRC collection of external scientific reports to inform the implementation of the EU AI Act on general-purpose AI models

The European Commission’s Joint Research Centre (JRC) has released a comprehensive collection of six external scientific studies on general-purpose AI (GPAI) models, marking a major step in the scientific groundwork for informing the implementation  of the EU AI Act. The collection presents methodologies and technical insights that will help identify, classify, and assess compute, capabilities and risks of GPAI models operating within the European regulatory framework.

The EU AI Act, which entered into force on 1 August 2024, aims to promote trustworthy innovation in artificial intelligence while safeguarding health, safety, and fundamental rights, including democracy and the rule of law. Chapter V of the Act introduces obligations for providers of GPAI models and for GPAI models with systemic risk, to be applied from August 2025. These obligations require precise definitions and assessment methods—an area where the JRC’s expert-driven research can help inform their implementation.

Between September 2024 and June 2025, the JRC established and coordinated a pool of worldwide experts to develop practical methodologies for categorising AI models as GPAI and determining when they might pose systemic risks. This work was carried out in close collaboration with the European Commission’s AI Office and other relevant services. It also informed the recently published Commission guidelines on the scope of obligations for GPAI providers, prepared through a public multi-stakeholder consultation.

The six reports in the series explore different aspects of the technical and scientific challenges involved in assessing  GPAI models:

• Training Compute Thresholds – Key Considerations for the EU AI Act
• A Framework for General-Purpose AI Model Categorisation
• A Framework to Categorise Modified General-Purpose AI Models as New Models Based on Behavioural Changes
• A Proposal to Identify High Impact Capabilities of General-Purpose AI Models
• The Role of AI Safety Benchmarks in Evaluating Systemic Risks in General-Purpose AI Models
• General-Purpose AI Model Reach as Criterion for Systemic Risk

Together, these studies provide a multi-dimensional perspective on how AI models can be evaluated in terms of compute power, generality, capability, behaviour, reach, and risks. The findings are expected to inform ongoing policy and regulatory work under the AI Act,  as well as stimulate future scientific discussions around the holistic evaluation of general-purpose AI models.

By combining scientific evidence with expert consultation, the JRC continues to underpin the EU’s approach to trustworthy AI, contributing to an implementation of the AI Act that balances technological progress with ethical and societal safeguards.