What happened
XBOW demonstrated that supplying a malicious URL as an Azure resource identifier caused the MCP server to make an authenticated SSRF request using its managed identity, leaking the token to the attacker-controlled endpoint.
2026-03-10
HighPrimaryFirst CVE targeting the MCP protocol layer in a production Microsoft system. A malicious URL in place of an Azure resource identifier caused the MCP server to make an SSRF request leaking the managed identity token, enabling AI agent privilege escalation. CVSS 8.8. Discovered by XBOW.
What happened
XBOW demonstrated that supplying a malicious URL as an Azure resource identifier caused the MCP server to make an authenticated SSRF request using its managed identity, leaking the token to the attacker-controlled endpoint.
Why it matters
Managed identity token exposure enabling privilege escalation in Azure environments via AI agent. Potential access to any Azure resource accessible to the managed identity — storage, databases, secrets, and other cloud resources.
Missing authorization check
MCP servers must validate that resource identifiers are legitimate, in-scope endpoints before using production credentials to make requests. Managed identity tokens must never be used to authenticate requests to attacker-controlled URLs.
Would PP block it?
If managed identity usage required a PP receipt bound to a specific authorized resource, attacker-supplied resource URLs would not match the authorized resource and the credential use would be blocked.
Incident analysis
2026-03-10
Microsoft Patch Tuesday: CVE-2026-26118 patched. First CVE targeting MCP protocol layer in a production Microsoft system. CVSS 8.8. Discovered by XBOW.
Authorization boundary
This incident is categorized as Credential exposure. The relevant Permission Protocol gate is Credential Gate. The read is conditional: the block only applies where the real action boundary is routed through a gate.
PP's Credential Gate would require authorization receipts before managed identity tokens are used for resource access. It would not prevent the SSRF URL injection — that requires URL validation in the MCP server code.
Related incidents and controls
Anthropic MCP Inspector Accepted Unauthenticated Connections Leading to CSRF-Chained RCE (CVE-2025-49596, CVSS 9.4)
nginx-ui MCPwn: Empty IP Whitelist Treated as Allow-All Grants Zero-Credential Access to All 12 MCP Tools on 2,600+ Exposed Instances (CVE-2026-33032, CVSS 9.8)
Start small
This incident maps to Credential Gate. Start with the boundary that controls the actual action, then require a signed receipt before execution.