By Hamza Faisal

A contractor working for Accenture reportedly connected a personal AI coding agent to a secure Pentagon production system, giving it administrative access to a network used by multiple intelligence agencies — and the agent ran there for about two months before anyone spotted it, according to a WIRED investigation published on October 29.

The system sits on NIPRNet, the US Department of Defense's network for unclassified information, and holds controlled unclassified information (CUI): material that is sensitive enough to require protection even though it is not classified. WIRED, citing sources familiar with the matter, reports that the contractor linked a personal account for Claude Code, Anthropic's AI coding tool, to the production environment earlier this year.

To automate coding work and cloud-server setup, the contractor handed the agent their own login credentials. That gave the agent administrative access, letting it bypass the security controls normally enforced through the login process — and allowing it to write and commit code on the contractor's behalf inside a live system.

Credentials and a network map sent outside

While doing its work, the agent appears to have transmitted the contractor's credentials, passwords and a map of the NIPRNet environment to an unnamed third-party website. WIRED did not name the site, saying that identifying it could expose the system to further risk, but reported that the National Institute of Standards and Technology had previously flagged it as high risk because of a vulnerability that could let attackers inject instructions into AI coding agents and steal credentials.

The report does not say that any attacker intercepted or used the information, and there is no indication that classified or top-secret systems were reached. The two-month window refers to how long the agent ran before it was identified and the vulnerability patched — not to confirmed data transfers throughout that period.

A reporting dispute

Department of Defense employees and contractors are required to report cybersecurity incidents within 72 hours of discovery. Sources told WIRED that Accenture did not initially report the incident to the government, and that its employees did not treat it as severe or urgent.

Accenture disputes the allegations outright. The company told WIRED that they were "thoroughly and independently investigated and found to be without merit" and that it "fully complied with our legal and contractual obligations." The Department of Defense declined to comment on the status of its networks or systems. The stakes are high: Accenture holds federal contracts worth more than $3.5 billion, including nearly $605 million with the Defense Department.

Why AI agents worry security teams

Claude is currently barred from use by Defense Department staff, and Anthropic pointed WIRED to its usage policy, which tells users not to use its products to interfere with the operation of military bases and related infrastructure, or to bypass security controls such as authentication systems without authorization.

The incident captures a new kind of security risk: an AI agent is only as safe as the credentials it is given, and a personal tool wired into a production system with standing admin access can quietly become a back door — sending sensitive data to outside services its user never inspected. As AI coding agents spread through government and enterprise work, the question is no longer just what the agent can do, but where everything it touches ends up.

By Hamza Faisal