liveBook · Manning

Open-source GPAI model

As explained in AI Governance

An open-source GPAI model is a general-purpose AI model made available under open-source terms whose provider may qualify for specified regulatory exemptions.

An open-source GPAI model receives only a partial exemption from EU AI Act documentation rules. Its provider must still publish a training-data summary using the AI Office template and maintain a copyright-compliance policy, but need not provide technical documentation on request. For high-risk use, missing provider documentation becomes a downstream responsibility; the exemption disappears entirely for models classified as systemic risk. section 7.3.5

Check these points before relying on the exemption:

  • Confirm the training-data summary and copyright policy are available. section 7.3.5
  • For high-risk use, identify documentation gaps and decide whether commercial support can supply more information. section 7.3.5
  • Check whether the model is classified as systemic risk; that classification removes the exemption. section 7.3.5
Figure 7.12 Llama 4’s model card on Hugging Face.

The model card illustrates published disclosures, including training-data sources, token counts, energy use, emissions, red-teaming methods and benchmark results; it does not detail licensing or copyright-clearance methods. section 7.3.5

How AI Governance covers it

How deep it goes
Explains disclosure duties and exemption limits, but stops short of tracing individual training-data licensing or clearance.
Examples
Uses Llama 4’s model card and commercial support examples involving Llama and Mistral.
What you can do after
Evaluate provider disclosures and decide whether support or internal documentation must close high-risk-use gaps.
How it connects
Connects open-source GPAI models to the EU AI Act, copyright compliance, high-risk use and systemic risk.

This explainer was generated by Manning's knowledge graph from the text of AI Governance. The section links point to the parts of the book it draws on.