An Accenture employee allegedly gave a personal Claude Code AI agent administrative access to a secure U.S. The agent reportedly operated for about two months and appears to have sent credentials and network details to an outside website. According to WIRED’s investigation, published October 29, the system serves multiple intelligence agencies and holds Controlled Unclassified Information, or CUI.
The employee allegedly connected their personal account for Anthropic's Claude Code to a production environment on the non-classified Internet Protocol Router Network, known as NIPRNet. Unlike classified military networks, NIPRNet connects to the internet while supporting protected government work. Wired withheld the site's name but reported that NIST identified a high-risk vulnerability that could allow attackers to inject instructions into an AI coding agent, potentially stealing credentials.
The issue is related to indirect prompt injection attacks, where hidden commands in external content trick an agent into performing unauthorized actions. Security experts informed WIRED that merging separate pieces of unclassified information can reveal sensitive details that warrant classification. Former NSA hacker Jake Williams highlighted another potential risk: stolen credentials could enable attackers to alter software developed on NIPRNet before it moves to more sensitive environments.
The publication highlighted Accenture's federal contracts, including nearly $605 million with the Defense Department, emphasizing the importance of oversight in government work.
