Where are tags applied to control access to Generative AI when implementing AI Access Security?

AI Access Security extends Prisma Access ' s existing App-ID-based application classification model to the generative AI space, and the mechanism it uses to let organizations differentiate their risk tolerance across the rapidly growing number of AI applications in use is to apply status tags — sanctioned, tolerated, or unsanctioned — directly to the identified Generative AI applications themselves, mirroring the same governance pattern long used for SaaS Security application risk classification. Once an AI application carries one of these tags, Security policy rules and dashboards can reference that classification consistently across the environment, giving administrators a scalable way to express organizational policy (which AI tools are approved, which are tolerated with monitoring, and which are explicitly prohibited) without having to hand-build a separate access rule for every individual AI application discovered. This makes option A the correct answer, since the tag is applied at the application object level, not any of the other locations listed. Applying tags to Security rules (option B) inverts the actual relationship: rules reference the application ' s tag/classification, they are not themselves the object being tagged. Tagging user devices (option C) would conflate device posture management with application classification, which are separate control domains in Prisma Access. Tagging Generative AI URL categories (option D) misattributes the classification mechanism to URL Filtering category objects, when AI Access Security ' s sanctioned/tolerated/unsanctioned tagging is applied to the discovered applications themselves via App-ID, not to a URL category construct.