Claude, Codex, and Hermes installed unowned code inside corporate networks

THE NEWS

What happened

Security researchers discovered that AI coding agents like Claude, Codex, and Hermes are automatically installing unowned code on corporate networks by trusting misconfigured 'llms.txt' files. This vulnerability affects over 100 websites, including Fortune 500 companies, breaking the trust model where agents blindly execute vendor documentation as ground truth.

CONTEXT

Why it matters

Critical security alert: AI coding agents including Anthropic's Claude, OpenAI's Codex, and Nous Research's Hermes are inadvertently installing unowned code inside corporate networks. Security researchers discovered that over 100 websites use misconfigured llms.txt files to serve executable content. When AI agents visit these sites, they blindly trust the documentation as ground truth and execute the code automatically. This supply-chain vulnerability has already triggered 'phone-home' responses from major Fortune 500 companies within hours of deployment. The emerging convention meant for site summaries is being weaponized, breaking the current security guards that cover SaaS and cloud layers.

AT A GLANCE

Key facts

• Researchers scanned 6,214 domains belonging to defense contractors and Big Tech firms.
• Over 8,200 llms.txt files were found, with 120 pointing to unregistered code packages.
• AI agents including Claude, OpenAI's Codex, and Nous Research's Hermes executed the malicious code.
• A Fortune 500 company was confirmed to have 'phoned home' to attackers within an hour of the exploit being deployed.
• The vulnerability exploits the emerging llms.txt convention intended for machine-readable site summaries.