Sources say a contractor's personal AI agent ran on a Pentagon network for two months and sent out login details. Here is why any business should check its own AI use.
An Accenture employee's personal AI agent reportedly ran on a secure Pentagon network for two months, according to sources who spoke to Wired. The agent is said to have sent the employee's login details and a map of the network to an outside website. Accenture says the allegations were investigated and found to be without merit.
The employee is alleged to have linked a secure work environment to their personal Claude account, so an AI agent could write and deploy code. Claude is an AI assistant made by Anthropic. To automate the work, the employee reportedly gave the agent their login credentials. That gave it administrative access and let it get around the security checks of the normal login.
The agent needed the internet to do its job. In doing so, it appears to have sent the employee's credentials, passwords and a map of the network to a third-party website. Wired is not naming the site. The National Institute of Standards and Technology has previously rated it high risk, because a flaw could let hackers plant instructions in an AI coding agent to steal credentials.
The network is the Pentagon's system for unclassified information. This environment held controlled unclassified information, which is sensitive and not meant for public release. It can cover critical infrastructure and nuclear submarine propulsion. One source said no one knows what happened to the data after it left, including whether it was kept.
Contractors must report a cyber incident within 72 hours of discovering it. Sources say Accenture did not initially report this one to the government at all. The Pentagon declined to comment on its systems. Accenture holds more than $3.5 billion in federal contracts, nearly $605 million of them with the Pentagon.
Government contractors are the most obvious group. So is any business whose staff use AI coding agents with company logins, cloud servers or sensitive data. The danger here did not come from chatting with an AI tool. It came from handing it a login and letting it act on its own inside a protected system.
The steps in the alert box above are the place to start. They come down to keeping personal AI tools off work systems and limiting what any AI tool can reach.
Much is still unclear. The details come from anonymous sources, and Accenture did not say who carried out its investigation or dispute that an incident took place. Whether the stolen details were used or kept is unknown. One expert, Matt Triner, said contractors face pressure to work faster and cheaper, and that firms can be punished for reporting breaches, even though this kind of incident would need to be reported.
Never give an AI tool your work login or password, and keep personal AI accounts away from company systems.
A viral ChatGPT prompt turns what the chatbot remembers about you into a room image. Posting it publicly could give scammers clues to target you.
Read more: Viral ChatGPT room image could hand scammers clues about youTuring Award winner Richard Sutton says more computing power alone will not make AI truly smart. Big AI companies disagree, which matters for anyone betting on AI.
Read more: AI pioneer says piling on computing power won't make machines smartA user says Anthropic's Claude Code put his email in web requests, then scam messages arrived. The link is unproven, but it shows what AI tools can share.
Read more: User says Claude Code sent his email address to websites
