Defense Contractor Employee Connected Personal AI Agent to Secure Pentagon Network, Sources Say

Reports indicate that earlier this year, a contractor at global consulting firm Accenture allegedly connected a secure Department of Defense centralized network to a personal account for Anthropic’s AI coding agent, Claude Code. According to sources speaking to the press, the agent ran on the network for two months before being identified, allegedly sending system information and login credentials to a third-party website.

Network Access and Controlled Unclassified Information

The environment accessed was the Non-classified Internet Protocol Router Network (NIPRNet), which handles controlled unclassified information (CUI). While NIPRNet is unclassified, it contains sensitive details regarding critical infrastructure and agency operations. Security experts note that credential exposure could theoretically aid external actors in navigating the infrastructure or understanding system layouts.

Official Responses and Investigations

• Department of Defense: A DOD spokesperson stated that the department does not comment on network or system status.
• Accenture: An Accenture spokesperson stated that allegations were independently investigated and found without merit, maintaining that the firm complied with all legal and contractual obligations, though details of the investigation were not provided.

Security analysts emphasize that integrating AI coding tools without authorization introduces supply chain and architectural risks, as automated tools can generate code or handle sensitive data without direct human oversight of every transmitted packet.