Skip to content

Azure Arc Windows Agent CVE-2026-26117 Lets Low-Privilege Users Escalate to SYSTEM and Seize Cloud-Managed Identity

CVE-2026-26117, a local privilege escalation flaw in the Azure Arc Connected Machine Agent for Windows, allows any domain user on a managed host to escalate to SYSTEM and inherit the host's Azure managed identity — granting access to all Azure resources the machine identity can reach. Microsoft rated the flaw CVSS 7.8; patch immediately given Arc's growing enterprise footprint.

Article identity-access-management

A local privilege escalation vulnerability in the Azure Arc Connected Machine Agent for Windows — the software that enrols on-premises and multi-cloud Windows servers into Microsoft’s hybrid cloud management plane — allows an attacker with standard domain user credentials to escalate to SYSTEM-level privileges and, critically, to exfiltrate or use the machine’s Azure managed identity token. That token, once obtained, can be used to authenticate to any Azure resource that the Arc-enrolled machine has been granted access to.

The flaw, CVE-2026-26117 (CVSS 7.8, HIGH), was discovered by Cymulate researchers and disclosed to Microsoft’s Security Response Centre under coordinated vulnerability disclosure. A patch was issued in the Azure Arc Windows agent version 1.39.

The Vulnerability Chain

Azure Arc’s Windows agent runs a local service (himds.exe) that vends managed identity tokens to processes running on the machine. Access to this token endpoint is intended to be restricted to processes running at elevated privilege levels. However, CVE-2026-26117 involves a race condition and insufficient access control on a named pipe used for internal agent communication.

By racing the named pipe connection timing during agent service startup or a periodic metadata refresh cycle, an unprivileged user-mode process can gain a handle to the pipe before the access control list is properly applied. Through this handle, the attacker can request a managed identity access token from the Azure Instance Metadata Service (IMDS) proxy — the same token that legitimate SYSTEM-level services use to authenticate to Azure.

The exploit sequence documented by Cymulate:

  1. Attacker authenticates to the Windows host with any domain user account
  2. Monitors for agent service startup or metadata refresh events using system event tracing
  3. Races the named pipe to obtain an unprotected handle
  4. Issues a managed identity token request through the handle
  5. Receives a valid short-lived Azure AD access token scoped to the machine identity
  6. Uses the token directly against Azure Management API, Azure Key Vault, or any Azure resource the machine identity can access

Scope of Exposure

The practical impact extends well beyond local privilege escalation on the affected host. In typical enterprise Arc deployments:

  • Arc-enrolled servers are granted Azure RBAC roles — commonly Contributor on resource groups, Key Vault Secrets User, or Log Analytics Contributor — to enable Azure Monitor, Defender for Cloud, and Automanage integration.
  • A managed identity token for a high-value server (a domain controller, a secrets management server, or a backup target enrolled in Arc) provides Azure-plane access to potentially sensitive resources.
  • The token’s default lifetime is one hour, providing an attacker with a persistence-free but usable access window without triggering conventional credential theft detection.

Environments that have enrolled Arc on sensitive servers with broad Azure RBAC role assignments are at greatest risk.

  • Update Azure Arc Windows agent to version 1.39 or later — Microsoft has published the update through Microsoft Update and Arc policy-based automatic update. Verify via azcmagent version on enrolled hosts.
  • Audit Arc machine identity role assignments — in the Azure portal, review each Arc-enrolled server’s RBAC role assignments (Identity → Azure role assignments); revoke roles that exceed minimum operational requirements.
  • Implement Azure IMDS access logging — Arc managed identity token requests do not generate verbose audit events by default; enable Azure Monitor diagnostic settings on key vaults and resource groups to log getSecret and listKeys operations by managed identity callers.
  • Review Arc enrolment scope — evaluate whether high-sensitivity servers (domain controllers, PKI infrastructure, PAM systems) should be enrolled in Arc at all; the management benefit must be weighed against the expanded attack surface.
  • Harden named pipe permissions — for hosts that cannot immediately patch, consult Microsoft’s workaround guidance; temporary access control changes to the relevant himds named pipe can mitigate the race condition.

Share this article

Related Intelligence

🔑 IAM

CISA Confirms Active Exploitation of Windows Task Host Privilege Escalation CVE-2025-60710 — Four Public Exploits Available

A link-following flaw in the Windows Host Process for Tasks allows any local user to escalate to SYSTEM privileges. Patched in November 2025, CVE-2025-60710 has been confirmed as actively exploited — CISA added it to the Known Exploited Vulnerabilities catalogue on 13 April with a 27 April federal deadline. Four public proof-of-concept exploits are now freely available on GitHub.

#privilege-escalation +5
🔑 IAM

Microsoft Entra Agent ID Role Misconfiguration Enabled Full Tenant Takeover via Service Principal Hijack

A flaw in Microsoft Entra's Agent ID role assignment model allowed an attacker with low-level Entra access to hijack privileged service principals and achieve full tenant administrator rights. Microsoft silently patched the issue on April 9; organisations with agentic AI workloads or automation service accounts should audit role bindings immediately.

#entra-id +5
🔑 IAM

Microsoft Entra ID Entitlement Management SSRF (CVE-2026-35431, CVSS 10.0) — Cloud IAM Attack Surface Disclosed Before Silent Server-Side Fix

A perfect-score SSRF vulnerability in Microsoft Entra ID Entitlement Management allowed unauthenticated network-accessible exploitation of Microsoft's cloud identity governance platform. Microsoft patched it server-side with no customer action required, but the disclosure surfaces a structural question enterprise security teams need to answer: how do you monitor for exploitation of a vulnerability in infrastructure you don't control?

#entra-id +6