WIRED reports that an Accenture employee working as a contractor for the US Department of Defense allegedly connected a personal Claude Code account to a secure NIPRNet environment holding controlled unclassified information. The AI agent reportedly ran there for about two months before it was identified and the gap was closed. According to the report, it appears to have sent login credentials, passwords and a map of the network to a third-party site rated high risk. Accenture says the allegations were investigated and found to be without merit, and the department declined to comment.

What it means: Whatever the outcome of this particular case, the scenario it describes is one of security leaders' biggest fears today. Coding agents run commands, read files and talk to outside services, often with the rights of the user who launched them. When an employee uses a personal account, the organisation has no control over where data ends up, what logging exists and which tools the agent can call. That the activity reportedly went unseen for two months shows traditional monitoring systems were not designed to recognise how such tools behave. For any business, not only defence organisations, the issue is current: developers and analysts want agents because they make them more productive, and if there is no approved option they will use their own. The best response is to provide company accounts with clear rules, restricted network access and logging, along with technical measures that keep unknown tools out of sensitive environments. Contractors and suppliers must be covered by the same policies. In Europe, DORA and the NIS2 directive already require control over tools that touch critical systems.

Read the original on WIRED

Related service: Custom software

Related stories