What happened
Verified via NVD REST API. CVE-2026-110631 (published 2026-10-14, CVSS 9.8 v3.1 / 9.3 v4.0, CWE-1188): the Periphery agent's trust lists default to empty and fail open, so any keypair is accepted and the caller is treated as a trusted Core. CVE-2026-110005 (published 2026-10-13, CERT-CC / KB 456950) is the sibling command-injection bug in the same agent. Both affect the 'periphery'/'agent' component of the Komodo platform.
Why it matters
Komodo's Periphery agent runs on managed servers to execute deployments/containers; a fail-open authentication default turns any adjacent network host into a root shell on server fleets. For AI deployments this matters because Komodo-style agents increasingly sit next to model-serving and data infrastructure with broad permissions — and the paired command-injection flaw means even 'legitimate' low-privileged users can reach root. Configure core_public_keys/passkeys and update to the fixed agent.
Attack vector
Unauthenticated attacker presents any generated keypair to the Periphery agent (empty core_public_keys/passkeys fail open by default), then issues Periphery API requests for terminal sessions and container operations — arbitrary command execution on managed hosts, typically as root. CVE-2026-110005: authenticated low-priv user with Execute/Read+Logs permission interpolates shell metacharacters in a container name passed to sh -c in DestroyContainer/SearchContainerLog.
Affected systems
Komodo Periphery agent through 2.3.3 (auth bypass); Komodo 1.19.1–<2.3.0 (CVE-2026-110005 command injection)
Mitigation
Update Periphery to the fixed release and explicitly populate core_public_keys / passkeys (do not run with empty trust lists); restrict TCP 8120 to trusted Core addresses. For CVE-2026-110005 upgrade to >= 2.3.0. VulnCheck advisory + moghtech/komodo repo.