BluePhishing agent-blueprint consent can create attacker-managed identities

Proofpoint demonstrated a consent route in which an attacker-controlled multitenant Microsoft Entra agent blueprint can lead a consenting tenant to instantiate local agent identities under an attacker-retained manager credential.

An attacker must obtain appropriate consent for the blueprint, and effective delegated versus application permissions are bounded by the actual tenant grants. The blueprint manager can provision up to 250 child agent identities, which may not appear in ordinary application inventories. Consent, agent creation and effective data access are distinct events. Revoke the manager and inspect child identities, token grants and downstream access independently; a parent identity disappearing from an inventory is not proof that all descendants lost permission.

Proofpoint expressly reported a proof of concept, not a verified live BluePhishing intrusion or stolen customer record. Review consent policy, child-principal ownership, grant revocation and actual resource authorization separately.

Evidence reconciliation and effect boundary. The consent grants a management relationship whose descendants require their own inventory and revocation review. An administrator may notice the original blueprint application but miss locally created agent identities, while an investigator must still establish each child identity, token and effective resource privilege. A child existing does not prove it obtained delegated user authority. This remains a proof of concept rather than a verified compromise.