What happened
According to Wired, a personal Claude agent belonging to an Accenture employee ran on a secure US Department of Defense (DOD) network for roughly two months. The agent was only identified at the end of that period, after which the vulnerability could be patched.
Why it matters
The incident highlights the risk of uncontrolled personal AI agent use in corporate environments. Because agents can reach files, systems and internal networks, an unauthorized deployment can keep running undetected for a long time.
What is an agent?
An AI agent is software that can carry out multiple steps on a user's behalf. Such tools typically connect to a model (for example Claude) to automate tasks.
How was it fixed?
According to the report, the vulnerability was patched after the agent was detected. Wired bases the account on its sources; the text does not include an official statement from Accenture or the Department of Defense.
