When a single token was enough to shake trust in the cloud
Everything seemed under control in Microsoft environments. MFA enforced, conditional access rules applied, logs showing no anomalies. Yet, in the shadows, an invisible flaw in Entra ID made it possible to become a Global Admin in any tenant. No phishing required, no stolen passwords. Just a special token, mishandled by a legacy API, that granted full control. This vulnerability, now patched and tracked as CVE-2025-55241, shows how the entire trust model of cloud IAM can rest on a single design error.
The core issue: Actor tokens
Actor tokens are internal Microsoft mechanisms, designed to let services delegate actions transparently. Normally invisible to users, they function like master keys inside the infrastructure.
Security researcher Dirk-jan Mollema discovered that the Azure AD Graph API, an aging but still active service, did not properly validate the origin of Actor tokens. The result: an attacker could generate a token in their own low-privilege tenant and then use it to impersonate a Global Admin in a completely different tenant.
This flaw was the result of two problems combined: an overly powerful internal mechanism and a lack of validation in a critical API.
Why this vulnerability stands out
Exploitation triggered none of the usual signals. No sign-in logs, no MFA prompts, no conditional access enforcement. Even organizations with mature IAM practices would have been blind to it.
To grasp the scale of the risk, imagine: a token created in a test tenant could be enough to seize control of Exchange Online, SharePoint, or critical Azure resources belonging to another organization. With no direct traces, defenders could only rely on indirect anomalies, such as unexpected role assignments, new accounts, or suspicious access patterns.
A wake-up call for cloud IAM
CVE-2025-55241 is more than just another CVE. It is a warning about the hidden trust mechanisms inside cloud platforms.
As Mitiga points out, Microsoft is not alone. All major providers – Google, AWS, Okta, Salesforce – rely on undocumented service-to-service delegation methods. They are rarely reviewed, often assumed safe, but can become perfect weapons when validation fails.
What organizations should take away
Microsoft patched the vulnerability quickly in July 2025, and there is no evidence of exploitation to date. But the lack of historical visibility means no one can be entirely sure it wasn’t abused earlier.
For security teams, the key takeaway is clear: do not rely solely on built-in IAM logs and protections. Defense must shift toward post-exploitation anomaly detection. SOCs need to actively hunt for weak signals: unexplained role changes, suspicious new accounts, unusual SaaS integrations, or sudden deployments of cloud services.
IAM at the Heart of Cloud Trust
The CVE-2025-55241 vulnerability will be remembered as one of the most serious identity flaws ever revealed in the cloud era. Not because it was exploited, but because it exposed a structural weakness: our reliance on invisible delegation mechanisms.
For CISOs and IAM leaders, the lesson is simple: anticipate the next flaw. The question isn’t if another vulnerability of this scale will appear, but when. The only defense is a stronger IAM governance model, broader visibility, and detection capabilities that go beyond standard logs.
Source: Mitiga – Breaking Down the Microsoft Entra ID Actor Token Vulnerability: The Perfect Crime in the Cloud
Share:
External Attack Surface Management
See your exposure before attackers do. Know what's already compromised.
of successful breaches start from publicly exposed assets.
