Top AI tools including Claude, Codex, and Hermes installed suspicious code inside corporate networks

New research has revealed that top AI tools like Claude, OpenAI's Codex, and Nous Research's Hermes have been found to install suspicious code within corporate networks. Cybercriminals are now able to exploit unclaimed llms.txt references on 120 domains, allowing AI agents to install malware if they execute outdated or hallucinated documentation commands.

These ".txt" files, containing information about AI agents, can be found on the websites of defense contractors, Fortune 500 organizations, and big tech companies. Among the 8,265 of these ".txt" files, 120 were found to be pointing at non-existent packages and domains. Researchers registered some of these unclaimed names and hosted packages that would communicate with their servers when installed.

In less than an hour, a Fortune 500 company started pinging these malicious packages, with numbers soon growing to "a few dozen more." This demonstrates that cybercriminals can also carry out such attacks. To protect against this vulnerability, companies should clean up their documentation and restrict AI agents from treating it as executable instructions.

While AI agents treating documentation as executable commands may not be resolved soon, organizations should consider the risks when granting AI agents permission to execute commands.