Accenture Contractor Connected an AI Agent to the Pentagon's Secure Network and Leaked Data to a Third-Party Site
An Accenture employee gave Claude Code access to the US Department of Defense's internal network. The agent carried credentials out to a third-party site. The incident occurred on October 30, 2026.
On October 30, 2026, it became known that an employee of contractor Accenture, working at the US Department of Defense, connected an AI agent to the Pentagon's secure internal network. The agent, based on Claude Code (Anthropic's tool for writing and running code), collected data from a network used by several intelligence agencies and sent it to a third-party site. The credential leak was confirmed. Other controlled unclassified information (official data that is not classified but requires protection) was also put at risk.
The access scheme differed from the standard one. According to a Wired source, the employee gave the agent a login and password for the work environment so that it could write and deploy code. The same credentials allowed the agent to bypass security controls. The standard procedure assumes that access to a secure environment is granted through corporate mechanisms and is restricted by role; here the agent worked from a personal account. For the Department of Defense this is a separate story: Claude is currently prohibited for use by department employees due to the supply chain risk status assigned by Anthropic (an assessment of the threat that a contractor or product poses to the system).
In practice, the incident shows where control breaks down. An agent with access to an environment performs tasks autonomously and can contact external addresses if it is not restricted at the network level. In such a scheme, the login and password become a workaround past access policies: they are received not by a person but by a program, and checks designed for an employee do not trigger. According to a Wired source, "this puts all the data that is there at risk," and "someone could gain the ability to enter the secure environment unnoticed." The same source called the situation "a 100 percent national security threat."
What remains unknown: exactly what data went to the third-party site and what that site is. The note also does not indicate how many people or systems could have been affected and what measures were taken after the risk was discovered.