A contractor working for Accenture reportedly connected a personal AI coding agent to a secure U.S. Department of Defense network, potentially exposing login credentials and sensitive network information, according to an investigation by WIRED.
The incident involved Claude Code, Anthropic’s AI coding tool, which the contractor allegedly linked to a production environment running on the Pentagon’s Non-classified Internet Protocol Router Network (NIPRNet). The network is used for unclassified government information, including controlled unclassified information (CUI) that must not be publicly released without review.
Sources told WIRED that the AI agent operated on the network for about two months before the issue was identified and addressed. To automate coding tasks and cloud server setup, the contractor reportedly provided the agent with login credentials that gave it administrative access and allowed it to work beyond the normal sign-in process.
During its operation, the agent appears to have sent the contractor’s credentials, passwords and a map of the network environment to a third-party website that security authorities had previously identified as high risk. WIRED did not name the website, citing concerns that revealing it could expose the system to further risk.
The exposed information could be valuable to attackers even if they never obtained the underlying data. Security specialists told WIRED that network maps can help an intruder understand how systems are connected and identify important services or potential paths for further access.
There is no indication that attackers accessed classified or top-secret systems as a result of the incident. However, experts warned that malicious code introduced into an unclassified environment could create risks if it were later tested or deployed in more sensitive systems without adequate review.
The Department of Defense declined to comment on the status of its networks or systems. Accenture, meanwhile, told WIRED that the allegations had been thoroughly and independently investigated and found to be without merit, adding that it had complied with its legal and contractual obligations.
WIRED also reported that federal employees and contractors are required to report cybersecurity incidents within 72 hours of discovery. Sources said Accenture did not initially report the incident to the government, although the company did not provide further clarification about its investigation when asked by the publication.
The case highlights the security risks that can emerge when AI agents are given broad permissions inside sensitive environments. Unlike a chatbot that only generates suggestions, a coding agent can access files, run commands, change code and communicate with external services. If it receives powerful credentials, its actions can have consequences beyond the task a user originally intended.
Organizations adopting these tools need to limit agent permissions, keep credentials out of prompts and agent environments, monitor outbound connections, and require human approval before sensitive changes are deployed. These safeguards are especially important for contractors and third-party workers who may have access to systems operated by government agencies or large organizations.
If this article helped you, please consider supporting our work. Every small contribution keeps Abijita.com independent and running.
Source: WIRED.
