AI systems are requesting access, acting on behalf of users, and operating with credentials your security team never approved. Every new agent is an identity — with permissions, entitlements, and risk — but most organizations govern them with the same ad-hoc processes they use for service accounts: shared credentials, static roles, and no audit trail. Opal brings agent identities under the same policy framework, auditability, and enforcement as human identities, so your least-privilege posture doesn't have a blind spot shaped like an LLM.