Anthropic’s Claude Tag gives AI agents independent identities

Anthropic's Claude Tag introduces an agent identity model that gives the AI assistant a workspace- or channel-scoped identity with configurable permissions, tool and API access, and audit logging. Administrators can isolate identities per channel, control which connectors and repositories Claude can access, revoke permissions centrally, and review logs; direct messages use a user's personal identity and connectors. The report emphasizes access control, RBAC on Enterprise plans, and planned identity-aware controls combining agent and user permissions to limit sensitive actions.