The risk tiers
Unacceptable risk: practices such as social scoring by public authorities or manipulative techniques, prohibited outright. High risk: systems listed in Annex III (employment, credit, essential services, law enforcement, critical infrastructure among them) and AI embedded as a safety component in products already regulated under Annex I. Limited risk: transparency obligations, for instance telling a person they are interacting with an AI or labeling generated content. Minimal risk: no specific obligation. General-purpose models carry their own documentation and, for systemic ones, evaluation duties.
Provider or deployer, the duties differ
A provider places a system on the market and bears the design obligations: risk management system, data governance, technical documentation, logging, human oversight design, accuracy, robustness and cybersecurity, conformity assessment and registration. A deployer uses it under its authority and must use it as instructed, ensure human oversight, monitor it, keep the logs and, for some uses, assess the impact on fundamental rights. Most enterprises are deployers of several systems and providers of none, which shapes their inventory.
What a governance program does with it
The first artifact is an inventory of AI systems in use with their classification, because every downstream duty depends on it. The second is the link to the existing risk and control program: a high-risk system’s risk management, logging and oversight duties are controls with evidence, like any other, and its residual risk sits in the register. For embedded AI in regulated products, the obligations apply by 2 August 2028, which puts the deadline inside most product roadmaps today.