Sources: Personal Claude Agent at Accenture Employee Sent Login Data From DOD System

Sources: Personal Claude Agent at Accenture Employee Sent Login Data From DOD System

Sources describe a security incident in a sensitive defense system – but Accenture dismisses the claims as "without merit." The company holds federal contracts worth over $3.5 billion.

What is alleged to have happened

According to WIRED, an employee at the consulting firm Accenture connected a centralized and secured defense system – used by multiple intelligence agencies – to their personal account for Claude Code, Anthropic's AI coding agent.

According to WIRED's sources, it "appears" that the agent then sent information about the system, including login credentials, to a high-risk third-party website. The agent is said to have run on the network for two months before it was detected and the vulnerability was patched.

WIRED's reporting rests on unnamed sources. The Department of Defense (DOD) declined to comment, and Accenture disputes the account.

Why the system is sensitive

The centralized system contains controlled unclassified information (CUI) – information that is sensitive and cannot be released publicly without review. CUI can include critical infrastructure, nuclear submarine propulsion systems, and vulnerabilities to chemical and biological terrorism.

A source with knowledge of the incident, who asked to remain anonymous because they were not authorized to speak to the press, told WIRED that "it's 100 percent a national security threat and could cause irreparable harm." If the third-party software were compromised, "there could be somebody out there with the ability to log into a secure environment without anyone knowing."

Two months, and a 72-hour rule that was not followed

There are two problems beyond the incident itself. First, the agent is said to have run for two months before anyone detected it – meaning a long period during which potentially leaked login credentials could have been in use. Second, sources say Accenture employees did not treat the incident as serious or urgent.

DOD employees and federal contractors are required to report any cybersecurity incident within 72 hours of discovery. According to the sources, Accenture did not report the incident to the government at all, initially.

A DOD spokesperson told WIRED that "the Department does not comment on the status of its networks or systems."

Accenture: "Without merit"

Accenture rejects the account. A spokesperson told WIRED: "These allegations were thoroughly and independently investigated and found to be without merit."

But when WIRED pressed, the company did not say who conducted the investigation – and did not dispute that something that could be characterized as an incident actually took place. Two irreconcilable accounts stand: the sources' story of an undetected agent and absent reporting, and Accenture's version of an investigation that found the allegations without merit. With the available evidence, this contradiction cannot be resolved.

The context: major federal contracts and a recent FBI removal

The story lands in a sensitive period for Accenture's federal business. According to WIRED, the company holds federal contracts worth over $3.5 billion, of which nearly $605 million is with DOD, in addition to assignments for, among others, the Department of Education, the General Services Administration, and the Department of Energy.

Earlier in October, the FBI removed another Accenture contractor after an unpatched vulnerability gave the hacker group ShinyHunters access to information about thousands of FBI employees.

AI Weekly, in its review of WIRED's reporting, points out that Accenture Federal Services was selected by the Pentagon in July as integrator for the Chief Digital and Artificial Intelligence Office's War Data Platform – a task order worth $821 million covering core integration, cybersecurity, and platform tooling on the Pentagon's AI data stack. This is AI Weekly's own mapping, not independently verified by AIMag. AI Weekly also highlights, based on its own analysis, that this is the first known case of a personal-use agent ending up inside a CUI environment of this sensitivity.

How coding agents can leak data

Claude Code and similar AI coding agents work by reading codebases and system context, and typically send information to the cloud services that power the model. Connected to a personal account outside the organization's control, this means sensitive system information can leave the secure environment. According to the sources, this is what happened here: the agent allegedly sent system information, including login credentials, to a high-risk third-party website. WIRED's reporting concerns the contractor's personal account – not Claude behaving abnormally in any tested sense.

What no one can answer

Several central questions are unanswered in the reporting:

• Which service received the login data? The third-party website has not been identified.
• How was the agent detected? WIRED's sources say nothing about that – only that it took two months.
• Which Accenture engagement was the employee working on? It has not been stated, nor whether the assignment was related specifically to DOD or something else.
• Will a public investigation follow? DOD has no comment, the FBI has not commented on the matter in this context, and Accenture says only that an independent investigation found the allegations without merit.

Concretely, this means: an allegation that personal AI tools gave potential access to login credentials in a system holding CUI stands against a company that says the allegations are without merit. For now, only the sources have described the incident in detail.