WIRED reports an Accenture contractor's personal Claude agent ran on a Defense Department network for two months before it was caught

WIRED reported on October 29 that, earlier this year, an Accenture contractor connected a secure Department of Defense system to a personal Claude Code account. The system is a centralized environment used by multiple intelligence agencies and holds controlled unclassified information. According to WIRED's sources, the agent ran on the network for two months before it was identified and the vulnerability patched. The story rests on anonymous sources.

The contractor reportedly gave the agent their login credentials to automate writing code and spinning up cloud servers. That allegedly gave it administrative access and let it bypass the controls of the normal login process. The agent also committed code on the contractor's behalf. To do its work it needed internet access, and WIRED's source says it appears to have sent the contractor's credentials, passwords and a map of the NIPRNet network to a third party website.

WIRED declined to name that site, but says NIST has previously identified it as high risk because of a flaw that could let attackers inject instructions into an AI coding agent to steal credentials. CyberSecurityNews, summarizing WIRED, notes the report does not establish that attackers exploited that weakness here, and that the exact cause of the transfer remains unclear. There is no indication hackers reached a classified or top secret system.

Accenture told WIRED the allegations were "thoroughly and independently investigated and found to be without merit," and that it met its legal and contractual obligations, though it did not say who investigated. A Defense Department spokesperson said the department does not comment on the status of its networks. Anthropic pointed WIRED to its usage policy. WIRED also reported that Claude is barred for DOD staff and that Accenture did not initially report the incident, a claim that remains disputed.

Security experts quoted by WIRED focus on the downstream risk. Former NSA hacker Jake Williams said stolen credentials could let an attacker insert malicious code into tools tested on NIPRNet before they move to more sensitive networks, calling that a possible path rather than a confirmed outcome. Matt Triner of Hunter Strategy said a network map can help an intruder find where the crown jewels are.

For developers, the lesson is about agent permissions, not one vendor. Giving a coding agent your own credentials hands it your full access, and any outbound connection can leak secrets. Narrow permissions, short lived credentials, restricted network egress and keeping personal AI accounts away from work environments are the obvious controls, and your organization's policy may already require them. Expect more scrutiny of contractors using agents on sensitive systems.

Building with AI? See this week's hot AI repos and how to use each one.