AI Coding Assistants Found Installing Unowned Code in Corporate Networks
Security researchers have uncovered a concerning pattern where AI coding assistants are installing code packages that have no identifiable owner or maintainer within corporate environments.
The investigation found 227 install commands in corporate documentation that pointed to code nobody appears to own or maintain. The affected systems include Anthropic's Claude, OpenAI's Codex, and the Hermes framework, all of which have been used by enterprises to automate software development tasks.
The findings raise significant supply chain security concerns. When AI systems install unverified or abandoned packages, organizations expose themselves to potential vulnerabilities, malicious code injection, or dependency confusion attacks. Unowned packages lack security updates, proper maintenance, and accountability—critical factors for enterprise software infrastructure.
Security teams are now urging organizations to audit their AI-assisted development workflows and implement stricter controls over package installation permissions granted to AI tools. This incident underscores the broader challenge of maintaining security hygiene as AI systems become more deeply integrated into software development pipelines.