By role · General-purpose AI model provider
Providers of general-purpose AI models face a distinct set of duties: technical documentation for downstream integrators, a copyright policy and training-data summary, and for the most capable models, evaluation, adversarial testing, incident reporting and cybersecurity. These duties sit with the model provider even when the model is used inside someone else's system.
A model is not a system. The model provider documents the model and its training; the system provider who builds on it documents the system. Both chains of documentation have to exist and reference each other, which is why downstream documentation is a control of its own below.
Where an instrument defines a higher tier by capability or compute, the duties step up to evaluation, red-teaming, incident reporting and security. The controls for those duties are listed with the risk areas they address, so the evaluation report and the incident record can be planned together.
Sorted by how many of the duties on this page each control satisfies, so the ones worth building first are at the top. A control page lists every other duty it serves, in every jurisdiction.
| Control | Satisfies | Supports | Owner · frequency |
|---|---|---|---|
| Frontier model safety and security framework Policy | 5 | 4 | Head of AI safety · annual |
| AI incident management and regulatory reporting Process | 3 | 3 | Incident coordinator · continuous |
| Technical documentation, model cards and instructions for use Process | 3 | 1 | Product or model owner · at launch and on material change |
| Contractual allocation of AI duties across the supply chain Contractual term | 2 | 0 | Legal counsel · once per ai system |
| Synthetic content labelling and provenance marking Technical measure | 2 | 0 | Engineering lead · continuous |
| Adversarial and red-team testing for generative AI Technical measure | 1 | 3 | AI security or safety lead · at launch and on material change |
| Accuracy, robustness, fairness and security testing Technical measure | 1 | 1 | Quality or testing lead · at launch and on material change |
| AI governance policy and accountability structure Policy | 1 | 0 | Executive sponsor for AI · annual |
| Training-data provenance and copyright register Process | 1 | 0 | Model development lead · at launch and on material change |
| Model release and change-management gate Process | 0 | 2 | Release manager · at launch and on material change |
Every published duty whose record names this audience. It is the recorded set, not every rule in the world; a jurisdiction missing here may simply not be mapped yet (open gaps).
Informational only, not legal advice. Verify every claim against the linked official sources and consult qualified counsel before acting.
