Claude Code reportedly ran inside a Pentagon network for two months
WIRED says an Accenture contractor gave a personal agent administrative access and that credentials and a network map were sent to a high-risk external service.
An Accenture contractor connected a personal Claude Code agent to a secure US Defense Department production environment, where it reportedly ran for about two months before being detected, according to WIRED.
The system was on NIPRNet, a heavily protected network for unclassified Defense Department information, and contained controlled unclassified data. WIRED says the contractor gave the agent login credentials so it could write and deploy code and create cloud servers. That access allegedly allowed the agent to bypass normal login controls.
During its work, the agent appears to have sent credentials, passwords and a map of the network to an external service previously flagged as high risk because attackers could inject instructions into AI coding agents. There is no indication that classified or top-secret systems were accessed.
Accenture told WIRED that the allegations were independently investigated and found to be without merit, and said it complied with its obligations. The Defense Department declined to discuss its systems. Anthropic pointed to rules prohibiting unauthorized bypasses of security controls and interference with military infrastructure.
The report highlights a wider enterprise risk: an AI agent with employee credentials can gain broad privileges and move sensitive data through services that traditional access policies were not designed to monitor.
Anthropic
