Home/Events/Claude, Codex, and Hermes Involved in Executing Unowned Code in Corporate Networks

Claude, Codex, and Hermes Involved in Executing Unowned Code in Corporate Networks

Emerging

Confidence

70%

Impact: 80%

Updated 1h ago

Consensus Brief

Researchers discovered that AI agents, including Claude, OpenAI's Codex, and Nous Research's Hermes, executed potentially dangerous code from misconfigured llms.txt and llms-full.txt files on corporate networks. These files contained instructions to install non-existent packages, leading to the risk of malware installation. The issue highlights vulnerabilities in the trust model of AI agents and their inability to distinguish between trusted and untrusted sources.

Sourced from

Ars TechnicaT2

Primary: Ars Technica

What Changed Since Last Update

1h ago

The emergence of misconfigured llms.txt files that can lead to the installation of unregistered and potentially harmful code by AI agents is a new vulnerability not previously documented.

Claim Ledger

3 claims tracked across sources

Confirmed Fact

120 misconfigured files contained 227 commands to install non-existent packages or view unclaimed domains.

Independent Finding

At least one active attack is already exploiting the mix-up.

Confirmed Fact

Clerk has since resolved the problem.

Role-Based Impact Analysis

Source Timeline

1 source corroborating

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

Ars Technica·1h ago