Why enterprise AI agents could become the ultimate insider threat

The article warns that proliferating agentic AI in enterprises dramatically increases attack surface and insider-like risk, citing multiple vulnerabilities and incidents (ServiceNow impersonation CVE, Amazon Q extension token compromise, OpenAI Codex CLI command-execution risk, and a hiring-bot data exposure) and summarizes OWASP-style mitigations—treat agents as first-class identities, apply least privilege and short-lived tokens, require step-up authentication, authenticate inter-agent communication, centralize monitoring and revocation, and limit agent autonomy and blast radius.