The European Commission's new AI Office has published its Guidelines for Providers of GPAI Models… these clarify that you only become the GPAI provider if there is "significant change in the model’s generality, capabilities, or systemic risk" rather than some fine-tuning
The key points: The European Commission’s new AI Office has issued Guidelines clarifying what constitutes a provider of a General Purpose AI (GPAI) model. An entity becomes a GPAI provider only if significant changes occur in the model’s generality, capabilities, or systemic risk, rather than for routine fine-tuning.
Why this matters: Clarification reduces uncertainty for developers and companies using AI models, helping define regulatory responsibilities and compliance obligations. It provides a clear threshold for when regulatory oversight applies, supporting innovation without excessive administrative burden.
What might happen next: AI developers will need to assess their activities against these guidelines to determine regulatory status. Regulatory bodies may refine enforcement and monitoring based on this clearer definition, which could impact investment and development strategies.
ADDITIONAL INFORMATION:
Guidelines carefully and publishing detailed analysis in Oliver's regular newsletter, Enterprise AI Governance, in the near future. At first glance, some key points stand out:
- The AI Office builds on its definition of GPAI models, clarifying that they are those trained with computational resources greater than 10^23 floating point operations and capable of generating language (text or audio), text-to-image or text-to-video.
- When modifying a GPAI model (e.g., fine-tuning), you only become the provider if there is "significant change in the model’s generality, capabilities, or systemic risk".
- Although the European Commission (AI Office) cannot legally take any enforcement action until 2 August 2026, the GPAI model provider obligations are nonetheless applicable from 2 August 2025. Signatories of the GPAI Code of Practice will be treated favourably.
Link to the Commission's Guidelines: digital-strategy.ec.europa
See also HR Policy Global recent article on: EU Publishes Voluntary Guidelines for Advanced AI Models amid Company Pushbacks
