Accenture contractor allegedly linked Pentagon system to Claude AI agent for two months
An Accenture employee connected their secure Pentagon system to a personal Claude AI agent, allowing it to write and deploy code for two months before detection. The system contained controlled unclassified information (CUI), including details about critical infrastructure and nuclear submarine propulsion. Security experts warn this exposed login credentials and network maps to a high-risk…
Key points
- Accenture employee allegedly linked Pentagon’s NIPRNet to personal Claude AI agent for code automation
- Incident went undetected for two months before patching, despite CUI exposure risks
- Experts warn exfiltrated credentials could enable backdoor attacks on classified networks like SIPRNet
The incident occurred on the Non-classified Internet Protocol Router Network (NIPRNet), a Pentagon network for unclassified data. Accenture claims the allegations were investigated and found without merit, but sources say the company did not initially report the breach to the government. The contractor, who cited ongoing conflict with Iran, argues the incident was mishandled. Accenture has over $3.5 billion in federal contracts, including $605 million with the Department of Defense. Experts warn such breaches could enable backdoor attacks on higher-security networks like SIPRNet or JWICS.
A Defense Contractor Gave Their Personal AI Agent Access to a Secure Pentagon System
wired.com · 29 October 2026
Loading the full article…
This text was published by wired.com and written by Vittoria Elliott. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
Coverage and discussion
1source- Hacker News discussion · 7 pointsnews.ycombinator.com
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Policy & Regulation
All →- Reddit to test login-only access in some markets this quarter · 1 src
- OpenAI updates Model Spec to prioritize human intent, safety and transparency · 4 src
- Nvidia CEO Huang allegedly linked to Singapore exec accused of feeding AI chip intel to China · 5 src
- German courts hold Google liable for AI Mode answers that mix up facts · 1 src
- Anthropic urges California to block unsafe AI models via court orders · 4 src
Comments
via GitHub Discussions