'It’s 100 percent a national security threat': Accenture contractor connected an AI agent to the Pentagon’s secure network, and then sent data to a high-risk website

• An AI agent was connected to a secure internal Defense Department network
• The agent was given login details to allow it to perform coding tasks, but also allowed it to bypass security controls
• The agent sent sensitive information, including the credentials, to a high-risk third party website

An Accenture contractor at the Department of Defense is in hot water over an AI agent that was given access to the Pentagon’s secure internal network.

The Claude Code agent then collected data from the network - which is used by multiple intelligence agencies - and sent it to a third-party website.

Login credentials were confirmed to have been leaked by the agent, with other controlled unclassified information potentially put at risk.

Don’t connect unauthorized agents to secure networks

According to a source familiar with the matter who spoke to Wired, “It puts all the data that’s in there at risk. There could be somebody out there with the ability to log into a secure environment without anyone knowing,” the source said.

“It’s 100 percent a national security threat and could cause irreparable harm.”

Despite Claude currently being banned from use for Defense Department staff due to Anthropic's supply chain risk designation, the Accenture employee is alleged to have given their personal Claude account access to a secure production environment in order to write and deploy code.

For two months, the AI agent had access to the secure internal network before the risk was identified and patched - but sources say once the connection was discovered it was not treated as severe or urgent by Accenture employees.

The contractor is alleged to have given the AI agent login credentials to the DoD’s Non-classified Internet Protocol Router Network (NIPRNet) in order to allow it to write code. But in order to function properly, the AI agent also needed the internet to perform its tasks.

The login credentials give the AI agent the ability to bypass the security controls enforced by the login process, and the AI agent then sent the credentials and a map of NIPRNet to a third party website.

“We are in a conflict with Iran right now,” the source added. “It’s the worst time to be having this kind of cavalier attitude.”

Both employees and contractors for the DoD are required to report any cybersecurity incidents within 72 hours, but according to the source, the Accenture employee did not raise the alarm.

“The Department does not comment on the status of its networks or systems,” the DoD told Wired.

“We take all concerns seriously. 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 said.