An Accenture contractor connected a personal Claude Code account to a secure Defense Department network, and the AI coding agent ran there for two months before the vulnerability was identified and patched, WIRED reports. Sources told WIRED the agent sent login credentials and information about the network to a third-party website that the National Institute of Standards and Technology has previously flagged as high risk.
The network, called NIPRNet, carries unclassified information, including controlled unclassified information (CUI). That category can include sensitive details about infrastructure, military equipment and hazardous materials. Sources said the contractor gave the agent login credentials and administrative access so it could write and deploy code. It needed an internet connection to do that work.
It is not clear what happened to the information after it reached the website. WIRED reports no indication that hackers accessed a classified or top-secret system. The Defense Department declined to comment on its networks. Accenture said the allegations were thoroughly and independently investigated and found to be without merit, and that it met its legal and contractual obligations.
Why could the exposure matter?
NIPRNet is separate from the military’s classified networks, SIPRNet and JWICS, which are not connected to outside networks. But security experts told WIRED that access to NIPRNet could still create risks beyond the data sent out. A stolen login could help an attacker enter the network, while details about how its systems fit together could help them find valuable targets.
Code built or tested on NIPRNet can also be used in more sensitive environments. Former NSA hacker Jake Williams told WIRED that malicious code might behave normally on NIPRNet and then cause harm if deployed on a classified network. There is no evidence that happened in this case. Standard code-scanning tools may not catch a deliberate backdoor because it is not necessarily a software flaw.
Reporting and oversight
Federal employees and contractors are required to report cybersecurity incidents within 72 hours of discovery. Sources told WIRED Accenture did not initially report this incident to the government, and that employees did not treat it as urgent. Accenture did not clarify who carried out its investigation. The company has more than $3.5 billion in federal contracts, including nearly $605 million with the Defense Department.
The incident comes as AI coding agents can produce code faster than people can review it. Williams told WIRED that makes it harder to check where code came from and what it does. The practical question for government agencies is how to use these tools without giving them access to systems or information they should not reach.
Why it matters
For government workers and contractors, the reported incident shows how giving a coding agent administrative access can expose credentials and sensitive network details, even on an unclassified system. WIRED found no indication that classified systems were accessed; what happened to the information sent to the third-party website remains unknown.
