A general-purpose AI model is designed to perform many kinds of tasks across applications and domains rather than serving one narrow function. Foundational large language models are a prominent example. section 10.30
Its broad usefulness also creates governance challenges: downstream providers need reliable information about capabilities and limitations, while powerful models may require stronger evaluation, risk mitigation, cybersecurity, and incident reporting. section 10.30
Current obligations for providers can include:
- Summarizing training data and documenting the model. section 10.30
- Publishing copyright-compliance policies and information for downstream integrators. section 10.30
- Providing evaluations, risk assessments, mitigations, cybersecurity protections, and incident-reporting processes for models classified as systemic risk. section 10.30
Figure 10.7 The four risk categories in the AI Act
Use the risk-category diagram to place general-purpose model obligations within the European Union’s wider AI risk framework. section 10.30
