A essential mixture of legacy parts may have allowed full entry to the Microsoft Entra ID tenant of each firm on this planet.
The deadly combine included undocumented tokens referred to as “actor tokens” and a vulnerability within the Azure AD Graph API (CVE-2025-55241) that allowed the tokens to work with any group’s Entra ID surroundings.
A risk actor exploiting the problem would have had entry to a slew of extremely delicate knowledge with out leaving any hint within the logs on the focused surroundings, aside from their very own actions.
Entra ID is Microsoft’s cloud-based identification and entry administration (IAM) service, previously often called Azure Energetic Listing (Azure AD), which gives organizations with single sign-on, multi-factor authentication, and safety controls throughout apps and sources.
A devoted Entra ID occasion represents a single group and manages safe entry to all of the apps used, each on-premise and cloud-based.
This may embody Microsoft 365 companies, customized and third-party SaaS merchandise like Salesforce, Dropbox, or cloud apps from Google, Amazon, or SAP.
Safety researcher Dirk-jan Mollema, founding father of offensive safety Outsider Safety, found a token validation flaw that gave him World Admin privileges in each Entra ID tenant.
This stage of entry permits full tenant compromise and opens the door to any service authenticated via Entra ID.
Impersonating any consumer within the tenant
In a technical weblog publish, Mollema explains that actor tokens are issued by a legacy service referred to as Entry Management Service, which “is used for authentication with SharePoint purposes and likewise appears to be utilized by Microsoft internally.”
The researcher discovered them whereas investigating hybrid Change setups. He observed that Change would request them when speaking with different companies on behalf of a consumer.
“The Actor token permits it to ‘act’ as one other consumer within the tenant when speaking to Change On-line, SharePoint and because it seems the Azure AD Graph” – Dirk-jan Mollema
Actor tokens aren’t signed, which means they can be utilized to impersonate any consumer within the tenant, and have a 24-hour validity with out the potential of being revoked throughout this era.
Mollema says that “this complete Actor token design is one thing that by no means ought to have existed,” as a result of they lack the correct required safety controls:
- there are not any logs when Actor tokens are issued
- since these companies can craft the unsigned impersonation tokens with out speaking to Entra ID, there are additionally no logs when they’re created or used
- they can’t be revoked inside their 24-hour validity
- they utterly bypass any restrictions configured in Conditional Entry
- we’ve got to depend on logging from the useful resource supplier to even know these tokens have been used within the tenant
The researcher says that Microsoft depends on actor tokens internally for service-to-service communication and that the corporate plans to take away them.
Microsoft calls them “high-privileged entry (HPA)” that enables an utility or service, “to impersonate different customers with out offering any proof of consumer context.
Whereas testing a number of methods to make use of an actor token, Mollema modified the tenant ID to at least one totally different from that producing the token, and despatched it to the deprecated Azure AD Graph API (graph.home windows.web), anticipating an “entry denied” message.
As a substitute, the error the researcher noticed indicated that the token was legitimate, however entry was unauthorized as a result of the identification of the consumer was not discovered within the tenant.
supply: Dirk-jan Mollema
Mollema tried once more, this time with a sound consumer ID from the focused tenant, and noticed Azure AD Graph API return the requested knowledge.
“I examined this in just a few extra check tenants I had entry to, to verify I used to be not loopy, however I may certainly entry knowledge in different tenants, so long as I knew their tenant ID (which is public data) and the netId of a consumer in that tenant.”
Utilizing the identical actor token, the researcher was in a position to impersonate the World Administrator within the goal tenant and carry out all of the actions related to the function (e.g. handle and create customers of various roles, modify configurations, reset passwords, add admins).
Mollema highlights that not one of the actions required for acquiring World Admin privileges generated any logs within the sufferer tenant.
From an attacker’s perspective, exploiting the problems would have been attainable in just a few steps, beginning with producing the actor token from a tenant below their management:
- Discovering the tenant ID for the focused surroundings may be achieved with public APIs primarily based on the area title
- Discovering a sound netId of a daily consumer within the goal tenant
- Crafting an impersonation token with the actor token from the attacker tenant utilizing the tenant ID and netId of the consumer within the sufferer tenant
- Itemizing all World Admins within the tenant and their netId
- Crafting an impersonation token for the World Admin
- Performing any learn/write motion via the Azure AD Graph API
Mollema notes that solely exercise within the final step can be recorded within the sufferer tenant.
You will need to word that Microsoft began the deprecation course of for the Azure AD Graph API service in September final yr.
In late June, the corporate warned that apps configured for prolonged entry however nonetheless used Azure AD Graph would now not be capable to use the APIs beginning early September 2025.
Mollema reported the problems to Microsoft on July 14 and the corporate confirmed that the issue was resolved 9 days later.
On September 4, Microsoft additionally patched CVE-2025-55241, describing it as a essential privilege escalation vulnerability in Azure Entra.
46% of environments had passwords cracked, practically doubling from 25% final yr.
Get the Picus Blue Report 2025 now for a complete take a look at extra findings on prevention, detection, and knowledge exfiltration developments.
