Everyone’s building AI agents. Almost nobody’s ready for what they do to identity.

The article warns that powerful AI agents can discover and exploit software vulnerabilities at speed and scale, undermining security models built on the assumption that a human is behind every login. It highlights risks from large-scale impersonation using stolen credentials, recommends treating AI agents like employees (cataloging agents, limiting permissions, requiring human approval for sensitive actions), moving to phishing-resistant authentication, and implementing continuous behavioral and identity verification to maintain accountability and reduce attack surface.