CVE-2026-63077

This is the second unauthenticated RCE in TeamCity's agent-facing protocol within a few years. CVE-2022-42417 and CVE-2026-63077 share the same root cause: the agent polling model requires the server to accept and process inbound requests before authentication completes. By design, the server must handle unauthenticated protocol messages — that's what makes zero-configuration agent deployment work across NATs and firewalls. The CVSS 9.8 and CISA KEV listing are accurate, but the EPSS of 0.00649 is conspicuously low. This likely reflects that TeamCity agent ports are rarely directly internet-exposed — agents phone home through firewalls. But the more plausible exploitation path isn't external attackers scanning for open agent ports. It's lateral movement from a compromised build agent. Once an agent is compromised, it has authenticated access to the server through the polling protocol, with permission to trigger builds, exfiltrate artifacts, and access secrets. That's the threat model that actually matters for mature CI/CD environments. The deeper question is whether JetBrains fundamentally redesigned the agent authentication handshake or just patched the specific message handler that was vulnerable. If it's the latter, expect another CVE in this class — the architectural permission to process untrusted inbound messages before authentication would remain intact, making different injection points in the same pre-auth path high-probability discovery targets. Audit your build pipeline architecture: confirm agent ports aren't internet-exposed, but also treat a compromised build agent as a trusted insider with server access. The risk isn't just the CVE — it's the protocol trust model that makes one compromised node a pivot point into your entire CI/CD infrastructure.

Reviewed through automated stages and approved by a human before publication.