Sources told WIRED the contractor, an employee of consulting firm Accenture, linked the system to a personal Claude account.
They said the agent sent information about the system, along with login credentials, to a high-risk third-party website, and that it operated inside the DOD network for two months before anyone noticed.
WIRED's reporting says pieces of that data, combined with material from connected systems, could be used to assemble classified information, including details on critical infrastructure, nuclear submarine propulsion, and chemical and biological terrorism vulnerabilities.
An anonymous source called it a national security threat that could cause irreparable harm.
Claude is also currently barred for DOD staff, so the connection broke another rule.
The agent was not only writing code but committing it, meaning saving it to the system on the employee's behalf, and it required internet access to some parts of the network that were meant to be air-gapped, or isolated from outside connections.
There is no indication that the contractor intended harm; the person was reportedly trying to use Claude to write and deploy code.
The DOD told WIRED it does not comment on the status of its networks. Accenture said the allegations were thoroughly and independently investigated and found to be without merit, though it did not say who conducted the investigation or dispute that an incident occurred. Sources said Accenture employees did not treat the matter as especially severe or urgent. The case raises questions about how personal AI agents are used by contractors working with government systems. It follows another episode earlier this month in which the FBI found an Accenture contractor had not patched a vulnerability that let the hacking group Shiny Hunters reach FBI data.
