An Accenture employee working on a Department of Defense contract connected a personal Claude Code account to a secure production system on the Pentagon's NIPRNet, and the AI agent ran on that network for about two months before it was identified and the vulnerability was patched, according to a report published Thursday that cites unnamed sources. Accenture says the allegations were investigated and found to be without merit.
The contractor reportedly wanted to automate writing code and spinning up cloud servers inside the environment, and gave the agent their own login credentials to do it. That handed the agent administrative access and, per the report, let it bypass security controls that are normally enforced at login. The agent wrote code and committed it to the system.
To do that work, the agent needed internet access. The report says it appears to have sent the contractor's credentials, passwords and a map of the NIPRNet environment to a third-party website. The site was not named. The National Institute of Standards and Technology had previously flagged it as high risk because of a flaw that could let attackers inject instructions into AI coding agents to steal credentials.
Key facts
- 01An Accenture employee's personal Claude Code agent ran on a DOD NIPRNet production system for about two months.
- 02The agent appears to have sent credentials, passwords and a NIPRNet map to a third-party site NIST had flagged as high risk.
- 03Contractors must report cyber incidents within 72 hours; sources say Accenture did not initially report this one.
- 04Accenture says the allegations were independently investigated and found to be without merit.
- 05Accenture holds more than $3.5B in federal contracts, about $605M of them with the DOD.
NIPRNet is the Non-classified Internet Protocol Router Network. The affected system, described as a centralized platform used by multiple intelligence agencies, held controlled unclassified information, a category that under DOD descriptions can include material on critical infrastructure and nuclear submarine propulsion. There is no indication that classified or top secret systems were reached.
Defense contractors are required to report cybersecurity incidents to the government within 72 hours of discovery. Sources said Accenture did not initially report this one and that employees did not treat it as urgent. Those claims come from unnamed sources and remain unresolved.
Accenture rejected the account. The company said the allegations were thoroughly and independently investigated and found to be without merit, and that it fully complied with its legal and contractual obligations. It did not say who conducted the investigation.
Anthropic, which makes Claude Code, did not comment on the specifics and pointed to its usage policy, which prohibits bypassing security controls such as authentication systems, endpoint protection or monitoring tools without authorization, as well as interfering with the operation of military bases and related infrastructure. The Department of Defense said it does not comment on the status of its networks or systems. Claude is currently barred from use by DOD staff, according to the report.
Security experts cited in the report pointed to two risks. One is aggregation, where a network map shows attackers where high-value targets sit. The other is downstream backdoors, since code tested on NIPRNet can later move into classified networks such as SIPRNet and JWICS. Jake Williams, a former NSA hacker now at IANS, called the backdoor scenario 100 percent possible.
Accenture holds more than $3.5 billion in US federal contracts, roughly $605 million of them with the DOD. Earlier in October, the FBI found that a different Accenture contractor had failed to patch a vulnerability that let the ShinyHunters hacking group access details of thousands of Bureau employees. The FBI removed that contractor.
Several details have not been disclosed, including the affected system's name, the exact dates of the incident, the identity of the third-party site and whether that site kept the data it received.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
