The New Insider Threat: Autonomous Systems With Excessive Permissions 

The report warns that autonomous AI agents and machine identities are becoming a major insider threat: over-privileged, unmanaged bots and service accounts can be hijacked or abused to perform destructive actions or wide-scale data exfiltration. It cites supply-chain and OAuth/token compromise incidents (including an Amazon Q VS Code extension supply-chain issue, Salesloft/Drift token abuse, and a Copilot "EchoLeak" flaw) as evidence, and recommends enforcing least privilege, continuous monitoring, software supply-chain hygiene, centralized identity inventory and lifecycle management, and human approval for high-impact automated actions.