BluePhishing Abuses Microsoft Entra Agent Blueprint Consent to Gain Tenant Access
Social Engineering & Fraud · 14 sources · Updated 6d ago · First seen Nov 9, 2026

Proofpoint researchers demonstrated BluePhishing, a proof-of-concept extension of OAuth consent phishing that abuses Microsoft Entra Agent ID to gain access to a corporate tenant. The technique lures a user into consenting to a malicious third-party agent blueprint, establishing a blueprint principal in the victim’s tenant. An attacker can then use blueprint credentials and an automatically assigned identity-creation permission to create tenant-local agents and obtain tokens carrying granted permissions, enabling persistent access and further exploitation. Proofpoint notified Microsoft before publication and reported no observed in-the-wild exploitation.
A single blueprint can support up to 250 agent identities, potentially multiplying the impact of one consent decision. Limited visibility into inherited permissions complicates governance and auditing, while low awareness of agent-consent risks may leave users more willing to authorize malicious AI agents. Proofpoint recommends restricting consent, enforcing least privilege, applying agent-specific identity controls, monitoring blueprint and agent creation events, and warning employees about agent-consent risks.


