A critical vulnerability has been published for Archon, the open-source AI coding agent. CVE-2026-111322 carries a CVSS score of 9.8 and affects versions 0.2.13 through 0.11.1.
The flaw is the simplest kind and the worst kind: the agent control API has no authentication at all, and by default it binds to 0.0.0.0 rather than localhost. Binding to 0.0.0.0 means listening on every network interface, so on a laptop joined to a café network, or a cloud instance with a permissive security group, that API is reachable by anyone who can route to the machine. With no auth in front of it, reaching it is the whole attack. It is classified as unauthenticated remote code execution.
What to do now
There is no vendor patch at publication. The mitigations are to firewall the port, or bind the service to 127.0.0.1 so it only accepts local connections. If you are running Archon on anything with a public IP, assume exposure and check before you patch.
The wider point: a coding agent runs commands, so an unauthenticated control API for one is not a data-exposure bug, it is a shell. This is the same class of risk that drove Perplexity to open-source its sandbox and Anthropic to move 150 engineers into security this month.
Sources:
The Hacker Wire ·
VulDB ·
Strix