---
title: BluePhishing: consent phishing for Entra AI agents. One click on a malicious agent blueprint lets an attacker mint up to 250 agent identities in your tenant | ctxsec
url: https://ctxsec.com/items/proofpoint-bluephishing-entra-agent-blueprint-consent-phishing-initial-access
description: "Proofpoint shows that Microsoft Entra Agent ID inherits OAuth consent-phishing risk. An attacker builds a multi-tenant agent blueprint in their own tenant with bundled permissions (requiredResourceAccess plus inheritable permissions) and sends a consent link. Once a victim consents, a blueprint principal appears in the victim tenant with an implicit AgentIdentity.CreateAsManager permission that can't be revoked from it; the attacker then uses the blueprint's own client secret to create agent identities (each needs a sponsor user's object ID, which is easy to find) and gets tokens for them with the granted permissions, for example User.Read.All and Group.Read.All via Graph. Proofpoint notes Microsoft documents that inherited permissions aren't visible in the portal or Graph, only in tokens, and cites Message Center MC1506733 saying regular users will be able to create agent blueprints by default from November 2026."
status: "research proof of concept (Proofpoint Threat Insight blog, Nov 9, 2026); Proofpoint says it has seen no in-the-wild use and notified Microsoft first"
published: 2026-11-13
---

> Proofpoint shows that Microsoft Entra Agent ID inherits OAuth consent-phishing risk. An attacker builds a multi-tenant agent blueprint in their own tenant with bundled permissions (requiredResourceAccess plus inheritable permissions) and sends a consent link. Once a victim consents, a blueprint principal appears in the victim tenant with an implicit AgentIdentity.CreateAsManager permission that can't be revoked from it; the attacker then uses the blueprint's own client secret to create agent identities (each needs a sponsor user's object ID, which is easy to find) and gets tokens for them with the granted permissions, for example User.Read.All and Group.Read.All via Graph. Proofpoint notes Microsoft documents that inherited permissions aren't visible in the portal or Graph, only in tokens, and cites Message Center MC1506733 saying regular users will be able to create agent blueprints by default from November 2026.

[← Context & Harness Attacks](/context) · [markdown](/items/proofpoint-bluephishing-entra-agent-blueprint-consent-phishing-initial-access.md)

research proof of concept (Proofpoint Threat Insight blog, Nov 9, 2026); Proofpoint says it has seen no in-the-wild use and notified Microsoft first

2026-11-13 11:45 AM ET

### BluePhishing: consent phishing for Entra AI agents. One click on a malicious agent blueprint lets an attacker mint up to 250 agent identities in your tenant

- **Why it's different from app consent phishing:** one blueprint can spawn up to 250 identities, agent consent controls are less mature, and autonomous agent activity blends in.
- **Defend:** keep user consent restricted (the same Enterprise apps consent settings apply to agent blueprints), use permission classifications, apply Conditional Access and Entra ID Protection for agents, and alert on audit events for agent blueprint and agent identity creation (they show as Add application / Add service principal).
- **Hunt:** a blueprint's application ID is the same in every tenant that consents, so a known-bad blueprint ID can be searched and removed across tenants. See the tip and the playbook for steps.

Sources: [Proofpoint: BluePhishing: Initial Access via Third-Party Agent Blueprint Consent (Nov 9, 2026)](https://www.proofpoint.com/us/blog/threat-insight/BluePhishing-Initial-Access-via-Third-Party-Agent-Blueprint-Consent)
