Skip to navigation

Threat reportVulnerabilityTL-2026-1986

CVE-2025-55241: Microsoft Entra ID Actor Token Flaw Allowed Cross-Tenant Global Admin Impersonation

criticalPATCHED

An undocumented service-to-service "Actor token" mechanism, combined with a tenant-origin validation flaw in the legacy Azure AD Graph API (graph.windows.net), let an attacker obtain an Actor token from their own low-privilege tenant and forge an unsigned impersonation JWT that any other tenant's Graph API accepted, silently impersonating any user including Global Administrators. First disclosed .

Key facts

IdentifiersTL-2026-1986, CVE-2025-55241
Also known asOne Token to Rule Them All, The Perfect Crime in the Cloud, Entra ID Actor Token Vulnerability
SeverityCRITICAL
CVSS9.8 (CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H)
First disclosed
Exploitation statusPublic proof-of-concept exploit for a linked CVE
EPSSCVE-2025-55241: 1.5% (72.0th percentile)
Affected productsMicrosoft Microsoft Entra ID (Azure Active Directory) — legacy Azure AD Graph API
Weaknesses (CWE)CWE-287
ToolsROADTools - S0684, roadrecon, roadtx
MITRE ATT&CK techniquesT1078 Valid Accounts, T1087 Account Discovery, T1199 Trusted Relationship, T1550 Use Alternate Authentication Material, T1552 Unsecured Credentials, T1590 Gather Victim Network Information
MotivationUNKNOWN
Detection rules9 (1 shown free)
Sources10 cited sources

Detection coverage

As of 2026-08-11, Threadlinqs Intelligence publishes 9 detection rule(s) for TL-2026-1986 across Splunk SPL, Microsoft KQL and Sigma, covering 8 indicator(s) of compromise. The whole corpus is readable without an account. Every threat comes with one complete detection in Splunk SPL, Microsoft KQL and Sigma, free for everyone; a free account adds daily TL Creds to unlock the rest. Paid plans add a larger daily allowance, correlation and the MCP server. Threadlinqs MCP server · View plans.

Free detection: Post-Exploitation Directory Dump, Cloud-Credential Persistence, and Trust-Policy Tampering Following Impersonated Global Admin Session

Detects the post-exploitation phase of CVE-2025-55241: once an attacker has an accepted forged impersonation token for a Global Administrator, they use tools like ROADtools' roadrecon component to bulk-dump the directory (users, groups, devices, BitLocker recovery keys) via the same anomalous InitiatedBy service/UPN pairing. Beyond enumeration, the researcher and Microsoft's own remediation guidance flag two higher-impact abuse classes under the same impersonated identity: (1) adding application/service-principal credentials (client secrets or certificate key-credentials) to establish durable cloud-native persistence that survives the CVE's own patch, and (2) modifying Conditional Access policy to weaken or bypass the tenant's authentication trust boundary. Matters because this is where the exploit converts from silent access into concrete data exposure, durable footholds, and long-term control-plane tampering.

  • Languages: Splunk SPL, Microsoft KQL, Sigma
  • ATT&CK: T1068, T1552, T1087, T1098, T1484
  • Data sources: azure_ad_audit_logs, entra_id_directory_audit, microsoft_365_management_activity_api
  • Severity: critical
  • Confidence: high

Splunk SPL

index=azure_ad sourcetype="azure:aad:audit" earliest=-24h
| spath
| eval svc_display=coalesce('initiatedBy.user.displayName','InitiatedBy.user.displayName')
| eval svc_upn=coalesce('initiatedBy.user.userPrincipalName','InitiatedBy.user.userPrincipalName')
| eval op=coalesce(operationName,OperationName)
| eval target=coalesce(targetResources,TargetResources)
| where svc_display IN ("Office 365 Exchange Online","Skype for Business Online","Dataverse","Office 365 SharePoint Online","Microsoft Dynamics ERP")
| where isnotnull(svc_upn) AND svc_upn!=""
| where match(op, "(?i)(read bitlocker|recovery key|list users|list devices|get device|read device configuration|read directory|add service principal credentials|certificates and secrets|add application key credential|update conditional access policy|add conditional access policy|delete conditional access policy|update policy)") OR match(target, "(?i)(bitlockerrecoverykey|alternativesecurityids|keycredential|conditionalaccesspolicy)")
| eval impact_class=case(match(op,"(?i)(bitlocker|recovery key)"),"credential_theft", match(op,"(?i)(service principal|key credential|certificates and secrets)"),"cloud_credential_persistence", match(op,"(?i)(conditional access|update policy)"),"trust_policy_tampering", match(op,"(?i)(list users|list devices|read directory)"),"directory_enumeration", true(),"other")
| stats count as event_count values(op) as operations values(target) as targets values(impact_class) as impact_classes earliest(_time) as first_seen latest(_time) as last_seen by svc_upn, svc_display
| eval first_seen=strftime(first_seen,"%Y-%m-%d %H:%M:%S"), last_seen=strftime(last_seen,"%Y-%m-%d %H:%M:%S")
| sort -event_count

Microsoft KQL

AuditLogs
| where TimeGenerated > ago(24h)
| where InitiatedBy has "user"
| extend InitiatedByDisplayName = tostring(InitiatedBy.user.displayName)
| extend InitiatedByUPN = tostring(InitiatedBy.user.userPrincipalName)
| where InitiatedByDisplayName in ("Office 365 Exchange Online", "Skype for Business Online", "Dataverse", "Office 365 SharePoint Online", "Microsoft Dynamics ERP")
| where isnotempty(InitiatedByUPN)
| where OperationName has_any ("Read BitLocker", "recovery key", "List users", "List devices", "Get device", "Read device configuration", "Read directory", "Add service principal credentials", "Certificates and secrets", "Add application key credential", "Update conditional access policy", "Add conditional access policy", "Delete conditional access policy", "Update policy") or tostring(TargetResources) has_any ("bitLockerRecoveryKey", "alternativeSecurityIds", "keyCredential", "conditionalAccessPolicy")
| extend ImpactClass = case(
    OperationName has_any ("BitLocker", "recovery key"), "credential_theft",
    OperationName has_any ("service principal", "key credential", "Certificates and secrets"), "cloud_credential_persistence",
    OperationName has_any ("conditional access", "Update policy"), "trust_policy_tampering",
    OperationName has_any ("List users", "List devices", "Read directory"), "directory_enumeration",
    "other")
| project TimeGenerated, OperationName, ImpactClass, InitiatedByDisplayName, InitiatedByUPN, TargetResources, Result
| order by TimeGenerated desc

Sigma

title: Post-Exploitation Directory Dump, Credential Persistence, and Trust-Policy Tampering via Impersonated Global Admin (Actor Token Abuse)
id: 4f8a1c92-6d3e-4b7a-9f2c-1a5e8d3b7c66
status: experimental
description: Detects bulk directory enumeration, BitLocker recovery key reads, application/service-principal credential (key/secret) additions, or Conditional Access policy modifications performed under the anomalous service-identity/UPN InitiatedBy pairing that indicates an active CVE-2025-55241 impersonation session — covering directory reconnaissance, cloud-credential persistence, and trust-policy tampering consistent with roadrecon-style directory dumping and Global Admin abuse.
references:
    - https://dirkjanm.io/obtaining-global-admin-in-every-entra-id-tenant-with-actor-tokens/
    - https://msrc.microsoft.com/update-guide/vulnerability/CVE-2025-55241
author: Threadlinqs Team
date: 2026-08-11
tags:
    - attack.privilege-escalation
    - attack.t1068
    - attack.credential-access
    - attack.t1552
    - attack.discovery
    - attack.t1087
    - attack.persistence
    - attack.t1098
    - attack.t1098.001
    - attack.defense-evasion
    - attack.t1484
    - attack.t1484.002
    - attack.tool.roadtools
    - cve.2025.55241
logsource:
    product: azure
    service: auditlogs
detection:
    selection_service:
        InitiatedBy.user.displayName:
            - 'Office 365 Exchange Online'
            - 'Skype for Business Online'
            - 'Dataverse'
            - 'Office 365 SharePoint Online'
            - 'Microsoft Dynamics ERP'
    filter_no_upn:
        InitiatedBy.user.userPrincipalName: null
    selection_sensitive_ops:
        OperationName|contains:
            - 'BitLocker'
            - 'recovery key'
            - 'List users'
            - 'List devices'
            - 'Read directory'
            - 'service principal credentials'
            - 'Certificates and secrets'
            - 'application key credential'
            - 'conditional access policy'
            - 'Update policy'
    selection_sensitive_targets:
        TargetResources|contains:
            - 'bitLockerRecoveryKey'
            - 'alternativeSecurityIds'
            - 'keyCredential'
            - 'conditionalAccessPolicy'
    condition: selection_service and not filter_no_upn and (selection_sensitive_ops or selection_sensitive_targets)
falsepositives:
    - Legitimate helpdesk BitLocker recovery-key retrieval, routine application-secret rotation, or a scheduled Conditional Access policy change performed through a normal (non-impersonated) admin session will not match the InitiatedBy service/UPN pairing precondition and should not trigger this rule; investigate any hit as high-confidence
level: critical

Known false positives:

  • Genuine helpdesk BitLocker recovery-key retrieval or bulk directory export via a real (non-impersonated) Global Admin session under normal support tooling — verify InitiatedBy is a real service pairing before escalating, not a routine admin action
  • Legitimate scheduled application-secret rotation or a sanctioned Conditional Access policy change ticket performed by IT operations under a genuine (non-impersonated) admin identity

Tuning notes: Chain this detection to TL-2026-1986-SPL-001/KQL-001/SIG-001 as a correlation search (same svc_upn within the same 24h window) to cut false positives and surface the full exploitation-to-impact chain in one alert. Treat any hit tagged impact_class=cloud_credential_persistence or trust_policy_tampering as the highest triage priority — these represent the attacker establishing a foothold or weakening the tenant's authentication trust boundary that will outlive the CVE's own patch window, and warrant immediate credential revocation plus a full Conditional Access policy diff against the last known-good baseline.

8 more detection rules (Splunk SPL, Microsoft KQL, Sigma) · unlock with TL Creds. Sign up free to unlock the detection rules or query them with the Threadlinqs MCP server (Purple)
  • Actor Token Impersonation — Microsoft Service Identity Paired with Real UPN in Entra ID Audit Logs · unlock with TL Creds
  • Post-Exploitation Directory Dump, Cloud-Credential Persistence, and Trust-Policy Tampering Following Impersonated Global Admin Session · unlock with TL Creds
  • Cross-Tenant Pivot via B2B Guest netId Harvesting and ROADtools Client Reuse Across Tenants · unlock with TL Creds
  • Actor Token Impersonation — Microsoft Service Identity Paired with Real UPN in Entra ID Audit Logs · unlock with TL Creds
  • Post-Exploitation Directory Dump, Cloud-Credential Persistence, and Trust-Policy Tampering Following Impersonated Global Admin Session · unlock with TL Creds
  • Cross-Tenant Pivot via B2B Guest netId Harvesting and ROADtools Client Reuse Across Tenants · unlock with TL Creds
  • Actor Token Impersonation — Microsoft Service Identity Paired with Real UPN in Entra ID Audit Logs · unlock with TL Creds
  • Cross-Tenant Pivot via B2B Guest netId Harvesting and ROADtools Client Reuse Across Tenants · unlock with TL Creds

Indicators of compromise

As of 2026-08-11, Threadlinqs Intelligence tracks 8 indicators of compromise for TL-2026-1986.

8 indicators of compromise · unlock with TL Creds. Compare plans

Technical analysis

Researcher Dirk-jan Mollema reported it to MSRC on 2025-07-14; Microsoft deployed a fix on 2025-07-17 and CVE-2025-55241 was published 2025-09-04.

CVE-2025-55241 ("Azure Entra ID Elevation of Privilege Vulnerability") stems from two components combined by security researcher Dirk-jan Mollema: undocumented "Actor tokens" issued by Microsoft's Access Control Service (ACS) for internal backend service-to-service (S2S) communication, and a tenant-origin validation gap in the legacy Azure AD Graph API (graph.windows.net).

Actor tokens are signed JWTs carrying a `trustedfordelegation: true` claim, issued with a 24-hour validity window, that let Microsoft services such as Exchange Online, SharePoint Online, Skype for Business Online, Dataverse, and Microsoft Dynamics ERP "act as" a user when calling other Microsoft backend services. Any authenticated party (including a low-privilege service principal in an attacker-controlled test tenant) could obtain one of these tokens from ACS. To impersonate a specific user, the caller wraps the signed Actor token inside a second, deliberately UNSIGNED JWT (`"alg": "none"`) containing a `nameid` claim holding the target user's internal "netId" identifier plus `aud`/`iss` claims naming the target tenant.

The flaw: Azure AD Graph API accepted this impersonation token as long as a valid netId existed in the named tenant — it never verified that the wrapped Actor token had actually been issued for, or by, that tenant. Mollema demonstrated this by changing only the tenant ID field in an impersonation token crafted in his own lab tenant; the API returned a "user not found" error (rather than an authentication/tenant-mismatch error) when using a wrong netId, confirming the token itself was accepted cross-tenant. Because netId values are small incrementing integers, they are brute-forceable in minutes to hours per tenant, and Global Administrator netIds could be enumerated the same way. A second pivot exists via Azure AD B2B: guest-user objects store their home-tenant netId in the `alternativeSecurityIds` attribute, so control of any tenant with guest users lets an attacker harvest home-tenant netIds and pivot outward, tenant by tenant.

Because the entire flow occurs directly between the attacker's client and Microsoft's backend token-issuance and Graph API layers, it bypassed MFA and Conditional Access entirely and generated NO log entries for the token issuance or use — the legacy Azure AD Graph API had no request-level audit logging. Only subsequent object modifications performed while impersonating a Global Admin would show up in Microsoft 365 audit logs, and even then with a telltale but easy-to-miss anomaly: the log entry's `InitiatedBy` shows a legitimate Microsoft service display name (e.g., "Office 365 Exchange Online") paired with the impersonated Global Admin's real UPN, rather than the service's own identity.

An attacker who fully exploited this chain could read or modify virtually anything in the victim tenant reachable via Azure AD Graph: user and group objects, Conditional Access policy configuration, application and service principal credentials/secrets, device records including BitLocker recovery keys, and by extension any Azure resource governed at the tenant level.

Mollema reported the flaw to MSRC on 2025-07-14; Microsoft confirmed a production fix on 2025-07-17, days before any public knowledge existed. A follow-up mitigation on 2025-08-06 further restricted Actor token issuance for Azure AD Graph to Microsoft-internal callers only, closing the service-principal-credential path. Azure AD Graph API was separately, and independently, retired globally on 2025-08-31. CVE-2025-55241 was formally published 2025-09-04, and Mollema published the full technical write-up on 2025-09-17. Microsoft states it found no evidence of exploitation prior to the patch, but the complete absence of token-issuance logging means defenders cannot retroactively verify this with certainty for the window before 2025-07-17.

MITRE ATT&CK techniques by tactic:

Privilege Escalation

T1078 Valid Accounts

Initial Access

T1078 Valid Accounts; T1199 Trusted Relationship

Discovery

T1087 Account Discovery

lateral-movement

T1550 Use Alternate Authentication Material

Credential Access

T1552 Unsecured Credentials

Reconnaissance

T1590 Gather Victim Network Information

Remediation

Patches

  • Microsoft deployed a server-side fix on 2025-07-17 correcting tenant-origin validation for impersonation tokens in the legacy Azure AD Graph API.
  • Microsoft added a further mitigation on 2025-08-06 restricting Actor token issuance for Azure AD Graph to Microsoft-internal service callers only, closing the service-principal-credential request path.

Immediate actions

  • Confirm the tenant-side Microsoft fix (deployed globally 2025-07-17) is in effect; this was a server-side platform fix requiring no customer action, but tenants should verify via Microsoft 365 admin communications.
  • Run the AuditLogs KQL hunting query published by Mollema to look for anomalous pairings of Microsoft-service display names (Office 365 Exchange Online, Skype for Business Online, Dataverse, Office 365 SharePoint Online, Microsoft Dynamics ERP) with real user UPNs in the InitiatedBy field, focused on the window before 2025-07-17.
  • Review Conditional Access policy changes, newly created privileged accounts, and application/service-principal credential additions made by Global Administrator accounts in the pre-patch window.
  • Audit B2B guest user configurations and external collaboration trust boundaries for signs of cross-tenant netId harvesting.

Workarounds

  • Decommission and block all remaining use of the legacy Azure AD Graph API (graph.windows.net), which was officially retired by Microsoft on 2025-08-31.

Longer-term hardening

  • Complete migration of any remaining applications, scripts, or automations off the legacy Azure AD Graph API (graph.windows.net) to Microsoft Graph — Azure AD Graph API was fully retired 2025-08-31.
  • Increase logging/SIEM coverage and retention for cloud identity provider service-to-service token issuance, not just object modification events.
  • Restrict and periodically review B2B guest user trust relationships and cross-tenant collaboration settings.

Affected and fixed versions

  • Microsoft — Microsoft Entra ID (Azure Active Directory) — legacy Azure AD Graph API
    Vulnerable versions: Azure AD Graph API (graph.windows.net) prior to the 2025-07-17 server-side fix
    Fixed in: Server-side fix live globally as of 2025-07-17; additional Actor-token issuance restriction 2025-08-06; Azure AD Graph API fully retired 2025-08-31

Timeline

  • Security researcher Dirk-jan Mollema reports the Actor token / Azure AD Graph tenant-validation flaw to the Microsoft Security Response Center (MSRC); MSRC opens a case the same day.
  • Microsoft deploys a server-side production fix correcting tenant-origin validation for impersonation tokens in the legacy Azure AD Graph API.
  • MSRC confirms to Mollema that the reported issue has been resolved.
  • Microsoft applies a further mitigation restricting Actor token issuance for Azure AD Graph to Microsoft-internal service callers only, closing the service-principal-credential request path.
  • The legacy Azure AD Graph API (graph.windows.net) is officially and globally retired by Microsoft.
  • CVE-2025-55241 (Azure Entra ID Elevation of Privilege Vulnerability) is formally published via MSRC.
  • Dirk-jan Mollema publishes the full public technical write-up, "One Token to Rule Them All," detailing the Actor token mechanism and exploitation chain.
  • Mitiga publishes its independent technical breakdown of the vulnerability, "The Perfect Crime in the Cloud."
  • Mitiga's blog post is updated with additional context, per the article's revision metadata.

Sources cited

Previous and next in vulnerability

Drafted by the Threadlinqs research pipeline from 10 sources; published ; last updated —.

Threadlinqs Intelligence — Real-Time Threat Detection Platform

// threat_feed
$ sort --newest
Showing all threats

Live intelligence console

Threat level

Latest Threats