Finding AI is the first control. The FI needs systems, policies and procedures that recognise AI use consistently in every relevant business and functional area, since governance and controls cannot attach to AI nobody has recognised; clear definitions, criteria and processes, supported by tools where useful, make this work. Where AI sits inside a third-party service or is used to deliver one (a software-as-a-service product with AI features, say), identification covers at least the services of material third-party providers. Roles are explicit, with one control function designated to own the identification systems and processes: business units may do the spotting, but the control function keeps enough independent oversight to ensure the process is applied consistently and catches new AI use over time, has the last word on whether something counts as AI, keeps clear records of the process and its outcomes, and reviews and refreshes the process for new technologies and for AI appearing in existing vendor services. Risk from AI that has not been identified must stay within appetite: where complete identification is impractical, because vendors do not disclose embedded AI or staff use shadow AI, the FI names the resulting risks and mitigates them, for example through clear staff rules on permitted and prohibited uses including embedded features and public tools, and through technical measures such as network monitoring and data loss prevention that detect unauthorised AI.
The graph holds this control, the 0 it maps to, and the evidence behind each claim, over MCP and REST.