MP.AIGOV.1055 Local AI agents should ask before acting and run in a trusted process
A local AI agent that approves its own actions can run commands, change files and reach systems with the user's access, with nobody confirming each step. One bad instruction, from a prompt or from content it reads, can do real damage on that device.
Remediation action
Turn off auto-approve in each agent's settings on the device, or block the agent with your endpoint management tools. Investigate agents whose host process isn't trusted.
Related links
%TestResult%