All controls

20 controls across five categories.

Export
Category
Type

Govern

Policy, roles, training and the risk management process. Ensures AI use in an organization has an owner and that a shared understanding of what is and isn't allowed exists and is understood within the organization.

GV.1The organization's position on AI, including the goals, opportunities and limits for its use, is documented.BaselineGV.2Roles and responsibilities concerning AI are assigned and documented.BaselineGV.3Personnel that uses AI is trained on its responsible, safe and effective use. Training content and attendance is recorded.BaselineGV.4AI risks are registered and reviewed at least annually as part of the organization's risk management process.Baseline

Comply

The legal and regulatory obligations that follow directly from using AI: knowing which laws apply, marking AI-generated content, and the extra steps required around higher-risk uses.

CM.1Relevant laws and regulatory requirements are known, documented and managed.BaselineCM.2AI-generated content and AI interactions are marked as AI when published or used externally.BaselineCM.3High-risk uses of AI are logged, human oversight is conducted, and affected subjects are informed.TriggerCM.4A Fundamental Rights Impact Assessment is performed before high-risk use of AI.Trigger

Register

What AI is deployed, for what purpose, and how proportionate it is to the task. The register is what makes the rest of the framework possible to check in the first place.

RG.1Deployed AI systems and providers are registered and classified.BaselineRG.2For each AI use, its purpose is registered. Where AI acts autonomously (agents, scripts or automations), what it does is described.Tier IIRG.3For each AI use, proportionality of the AI system to the task is weighed against documented alternatives.Tier IIRG.4At decommissioning, AI systems follow the organization's regular offboarding process. Access and accounts are removed, and the AI register is updated.BaselineRG.5AI risks and data exposure to training are considered in privacy assessments of new and existing suppliers.Baseline

Access

Access rights for AI systems themselves, not only for the people who use them. AI that acts with its own credentials or autonomously needs the same access discipline as any other account.

AC.1An access review is conducted before an AI system inherits users' access rights.BaselineAC.2AI systems with separate access rights are given a named identity, and these rights are limited to a documented scope.TriggerAC.3Each autonomous AI system is assigned a named owner who is accountable for its actions.TriggerAC.4Access rights for AI systems are granted according to the organization's access control model.Tier II

Track

How deployed AI actually performs once it's in use: effectiveness, internal feedback and errors, collected so the organization's understanding of AI risk and effectiveness is based on proven metrics.

TR.1Key performance indicators for deployed AI systems are established to measure their effectiveness.Tier IITR.2Internal feedback regarding deployed AI usage is collected and reviewed.Tier IITR.3Errors in AI output are logged and communicated to strengthen awareness of AI risk.Tier II