AI Coding Tools Found Installing Unowned Code Inside Corporate Networks
The short answer
Researchers discovered 227 install commands in corporate documentation generated by Claude, Codex, and Hermes — all pointing to software packages that have no legitimate owner. If you've hired developers who use AI coding tools — or if your team uses them directly — verify that any AI-suggested software installations are checked against an official package registry before running them.
- security
- AI coding tools
- supply chain risk
- Claude
- Codex
- small business IT
What happened
Security researchers examining corporate networks found that AI coding assistants — including Anthropic's Claude, OpenAI's Codex, and Hermes — had generated install commands pointing to software packages that have no legitimate owner. According to Ars Technica, 227 such commands were discovered embedded in corporate documentation. Because these packages don't exist under any verified publisher, anyone who runs those commands could expose their network to malicious code if an attacker registers the package name first.
The finding spans multiple widely used AI systems, suggesting it reflects a broader pattern in how these models recommend software dependencies — not a problem isolated to one vendor. The presence of these commands in corporate documentation means they may have already been distributed across teams and systems without anyone flagging the risk.
What it means for your business
If you've hired developers who use AI coding tools — or if your team uses them directly — verify that any AI-suggested software installations are checked against an official package registry before running them. Ask your developer or IT contact to audit recent AI-generated code and documentation for unverified package names. Do not execute install commands from AI-generated scripts without confirming the packages exist and are actively maintained by a recognized publisher. The risk scales with how much AI-generated code has been deployed in your systems: a quick audit now costs far less than recovering from a compromised network. Treat AI-written install instructions the same way you would code from an unfamiliar outside contractor — review before you run.
Source: Ars Technica