A Claude agent ran inside a Pentagon network for two months, WIRED reports

An Accenture contractor connected a sensitive Pentagon system to their personal Claude Code account. The AI agent then ran on a Defense Department network for two months, Vittoria Elliott reported for WIRED on Thursday 29 October, citing sources. The agent appears to have sent login credentials and a network map to a high-risk website, the report says.

Accenture denied wrongdoing. TNW has not independently verified the report.

“These allegations were thoroughly and independently investigated and found to be without merit. We fully complied with our legal and contractual obligations,” an Accenture spokesperson told WIRED.

Accenture did not say who carried out the investigation, WIRED reported. The Defense Department told WIRED that it does not comment on the status of its networks or systems.

What the agent did

The system holds controlled unclassified information (CUI), which is sensitive data that needs review before any release. Several intelligence agencies use it. It runs on NIPRNet, the Pentagon’s network for unclassified information. NIPRNet connects to the internet but has heavy security.

The contractor gave the Claude agent their login credentials so it could write code and set up cloud servers, a source told WIRED. That gave the agent administrative access and let it get around the normal login controls. The agent wrote code and saved it into the system for the contractor.

It also had to reach the internet to do its tasks. In doing so, it appears to have sent the credentials, passwords and a network map to a third-party website. WIRED did not name the site. The US National Institute of Standards and Technology has flagged that site as high risk, WIRED reported. A flaw there could let hackers steer an AI coding agent into stealing credentials.

“It’s 100 percent a national security threat and could cause irreparable harm,” a source aware of the incident told WIRED.

A breach that was not reported

Defense Department staff and federal contractors must report cybersecurity incidents within 72 hours. Accenture did not first report this one to the government at all, WIRED’s sources said. Its staff did not treat it as urgent. Pentagon staff currently may not use Claude, after the department labelled Anthropic a supply chain risk.

Anthropic pointed WIRED to its usage policy, which bars using its products to get around security controls without authorisation.

There is no sign that hackers reached a classified system, WIRED reported. But teams often test code on NIPRNet before it moves to classified networks, said Jake Williams. He is a former NSA hacker who teaches at IANS.

“The code could operate completely normally in NIPR and do something malicious and undisclosed in a SIPR network, that is 100 percent possible,” Williams told WIRED.

‘We have not seen this before’

WIRED’s editors discussed the scoop on its Uncanny Valley podcast on Friday 30 October. Leah Feiger, WIRED’s director of politics and science, said the incident was new.

“We have not seen this before,” Feiger said.

Contributing editor Zoë Schiffer noted that there is no sign of a hacking attempt. The contractor was trying to use Claude to write and deploy code, she said.

“We don’t know that anything nefarious was going on,” Schiffer said.

Executive editor Brian Barrett called it “inexcusable”, and said he assumed it would not be the last case of its kind.

Accenture holds more than $3.5bn in federal contracts, nearly $605m of them with the Pentagon, according to WIRED. Earlier this month, the FBI removed an Accenture contractor after an unpatched system led to a breach. Accenture also caps staff Claude spending at $29 a month, 404 Media reported in October.