As cited
Copy frozen at (site build).
ai security
AI Coding Agents Are Installing Unknown/Untrusted Code on Corporate Networks
Researchers discovered that artificial intelligence (AI) coding agents from vendors including Anthropic, OpenAI, and Nous Research automatically installed packages from unregistered domain names and code repositories found in llms.txt configuration files hosted on corporate networks. The team scanned over 6,000 domains belonging to defense contractors, Fortune 500 companies, and Big Tech firms, identified 120 files pointing to unclaimed names, registered some domains, and received phone-home signals from multiple Fortune 500 companies within hours of hosting test packages. The researchers warn that AI agents treat vendor documentation as authoritative without verification, creating a supply chain attack surface comparable to past incidents like SolarWinds.
Why it matters: Security teams deploying AI coding agents at defense contractors, Fortune 500, and Big Tech companies face immediate risk of compromise through malicious package injection via configuration files that AI agents blindly trust and execute without human oversight.
- Source published
- First seen by Cybersecurity Tracker