Your Next Insider Threat May Be an AI Coworker

**Executive summary:** This article warns that agentic AI desktop assistants can perform actions on users' machines (clicking, typing, opening files) and therefore introduce operational security risks—such as mass deletion or modification of SharePoint-synced files, prompt-injection-driven actions, and insufficient admin controls—and recommends least-privilege, scoped permissions, human approval for high-impact actions, and agent governance before enterprise adoption.