Vulnerability  ·  2026-10-17

Archon AI coding agent: unauthenticated /api/* control API (binds 0.0.0.0) → remote code execution via the coding agent running with bypassPermissions

VulnerabilityHigh impactGlobalCVE-2026-111322
CVE-2026-111322 (CVSS 9.8 v3.1 / 9.3 v4.0, CWE-306) verified via NVD REST API (published 2026-10-15 by VulnCheck, status Deferred): the /api/* control API of the Archon AI coding-agent server is missing authentication and binds to 0.0.0.0 by default, so any network-reachable attacker can inject messages into the coding agent, which runs with bypassPermissions, yielding host command execution as the Archon user.
A classic example of the emerging default-insecure agent-framework class: the thing meant to control the agent is itself unauthenticated and bound to all interfaces. Since the agent bypasses permission gating, an attacker needs no credentials at all to reach real command execution on the host plus steal agent secrets. Any exposed Archon install is effectively root-by-default.
Unauthenticated network access to the /api/* control API: attacker sends messages to the AI coding agent, which runs with bypassPermissions, executing commands on the host as the Archon user. Also enables credential/config extraction and agent manipulation.
Archon 0.2.13–0.11.1 (coleam00/Archon — the popular Dify-based AI agent builder framework)
Vendor CNA (VulnCheck) records version ≤0.11.1 as affected; restrict/authenticate access to the control API, do not expose 0.0.0.0:port to untrusted networks, and upgrade. See VulnCheck advisory.
NVD REST API | CVE-2026-111322VulnCheck advisoryResearcher writeup (hackmd)
See this in the live feed Explore related AI security and governance findings — updated every morning.
Open the feed →