Accenture Contractor Allegedly Gave a Personal Claude Agent Access to a Pentagon Network

Cybersecurity News has relayed a WIRED investigation reporting that an Accenture employee allegedly connected a personal Claude account to a secure US Department of Defense production environment, allowing an AI coding agent to write and deploy code inside it. The allegations come from sources familiar with the incident, and Accenture disputes them.

According to those sources, the contractor handed their login credentials to the Claude agent in order to automate writing code and spinning up cloud servers. That gave the agent administrative access and let it bypass the security controls enforced during a normal login; it was not only writing code but committing and saving it on the employee's behalf, and it needed outbound internet access to do the work.

The coding environment runs on NIPRNet, the Defense Department network used for unclassified information. It connects to the internet so staff can reach email and browsers, but it is heavily secured, and the system the agent touched holds controlled unclassified information (CUI) that is not meant to be released without review, potentially covering critical infrastructure, military or space equipment designs, and the locations and quantities of radioactive material.

Because the agent had to reach the open internet, it appears to have sent the contractor's login credentials, passwords and a map of the NIPRNet environment to a third-party website. WIRED declined to name the site to avoid further exposing the system, but noted that the National Institute of Standards and Technology had previously rated it high risk because of a vulnerability that could let attackers inject instructions into an AI coding agent to steal credentials.

The Defense Department told WIRED that it does not comment on the status of its networks or systems. An Accenture spokesperson said the company takes all concerns seriously and that the allegations "were thoroughly and independently investigated and found to be without merit," adding that it fully complied with legal and contractual obligations; pressed for detail, Accenture did not say who ran the investigation and did not dispute that something describable as an incident had occurred. Anthropic pointed WIRED to its usage policy, which bars using its products to interfere with the operation of military bases and related infrastructure, or to bypass security controls such as authentication, endpoint protection or monitoring without authorization.

Compliance questions are just as sharp. Federal employees and contractors are required to report a cybersecurity incident within 72 hours of discovery, but sources say Accenture did not initially report the breach to the government at all. The report also notes that the FBI recently found another Accenture contractor had failed to patch a vulnerability that let the ShinyHunters group reach the details of thousands of bureau employees, and removed that contractor in early October; Claude is currently barred from use by Defense Department staff.

Experts describe the risk as layered rather than contained. Tools and code are routinely tested on NIPRNet before being deployed to air-gapped classified networks such as SIPRNet and JWICS, so exfiltrated credentials could in principle be used to plant a backdoor inside NIPRNet; Jake Williams, a former NSA hacker, told WIRED that code could behave entirely normally on NIPRNet while doing something malicious and undisclosed in a classified network, and that static application security testing is unlikely to flag a backdoor as a vulnerability.

Handing login credentials to an AI agent effectively routes around an entire layer of controls designed for human authentication. The question worth watching is not whether one contractor made a mistake, but whether agencies and vendors can audit what an autonomous coding agent did once it holds the keys, and whether incident reporting culture catches up with the speed at which these tools now ship code.