Accenture Employee Allegedly Gave Claude Code Admin Access to Pentagon Environment

An Accenture employee allegedly connected a personal Anthropic Claude Code account to a production U.S. Department of Defense environment on NIPRNet, granting the AI agent administrator privileges to write and deploy code. According to reports citing WIRED’s October 29 investigation, the agent operated for roughly two months and appears to have transmitted credentials and network details to an unidentified external website. The environment reportedly supports multiple intelligence agencies and contains Controlled Unclassified Information; there is no indication that attackers accessed classified networks.

The cause of the alleged transfer remains unestablished. Indirect prompt injection was raised as a possible mechanism, but neither exploitation nor a downstream software compromise has been confirmed. Accenture denied the allegations, saying an independent investigation found them without merit and that it met its legal and contractual obligations; claims that it initially failed to notify the government also remain disputed. The reported incident highlights the need to restrict personal AI accounts in production environments, enforce least-privilege access, and monitor agents’ outbound communications.