WIRED reported on October 29 that an Accenture contractor connected a personal Claude Code agent, Anthropic’s artificial intelligence coding agent, to a Defense Department production system on the unclassified NIPRNet, where it ran for about two months before the exposure was found. The system holds controlled unclassified information, or CUI, and serves multiple intelligence agencies. The agent appears to have sent the contractor’s login credentials, passwords, and a map of the network to a third-party website that the National Institute of Standards and Technology has flagged as high risk. To automate coding and cloud-server setup, the contractor gave the agent their login credentials, a source aware of the incident told WIRED, handing it administrative access and letting it bypass security controls enforced at login. NIST flagged that site over a flaw that could let attackers inject instructions into an AI coding agent and steal credentials; no evidence shows the flaw was exploited here. Jake Williams, a former National Security Agency hacker and faculty member at the security consultancy IANS, told WIRED that separate pieces of unclassified information can become classified once aggregated. Defense contractors must report cybersecurity incidents within 72 hours, and Accenture did not initially report this one, the source said. Accenture said it independently investigated the allegations and found them without merit, though Claude is already barred from Defense Department work.