Critical Flaw in Archon Coding Agent Lets Strangers Take Over Your Machine
A maximum-severity-class bug (CVSS 9.8) in the Archon AI coding agent exposes an unauthenticated control interface. No patch exists yet; the interim fix is to lock the interface to your own computer.
A newly published vulnerability in Archon, an AI coding agent, scores 9.8 out of 10 on the standard severity scale (CVSS). It affects versions 0.2.13 through 0.11.1. In plain terms: the control panel of the agent can be reached by anyone on the same network without a password, and the agent behind it can run commands on the computer.
The details
According to The Hacker Wire, the agent’s /api/* control interface has no authentication and, by default, listens on 0.0.0.0. That address means “accept connections from any device that can reach me,” not just your own computer. The agent also runs with bypassPermissions, a setting that skips the usual “are you sure?” prompts. Combine the two and a stranger who can reach the port can send the agent instructions and have it run commands as you.
No vendor patch was available at publication. The report says there is no public proof of concept and the flaw is not on the US government’s list of actively exploited bugs. The article itself is an AI-assisted summary, so check the official entry before acting.
What is behind this
Coding agents are useful because they can act: edit files, run programs, install things. That same power is why a missing lock on the front door matters so much. Many early tools are built for a single developer on a laptop and quietly assume a friendly network. The trouble starts when someone runs them on a shared Wi-Fi, a cloud server or an office network.
What this means for you: If you do not use Archon, nothing to do. If you do, until a fix ships, restrict the port with a firewall or rebind the interface to 127.0.0.1, which means “only this computer,” and avoid running it on shared networks. More generally, before trying any agent tool, check which network address it listens on and whether it asks for permission before running commands.
Sources
Microsoft Previews MAI-Base-2, a 2.8-Trillion-Parameter Model It Will Not Open Up
Microsoft's MAI-Base-2 is a 2.8T-parameter foundation model trained on 50T tokens that Microsoft says learns far more efficiently than DeepSeek's. It is still in training and will go only to trusted partners.
