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

Anthropic withheld its advanced AI model Mythos due to risks of exposing critical software vulnerabilities, highlighting the dual-use nature of AI agents capable of both enhancing and compromising security. The article warns that AI systems, while automating tasks like scheduling and shopping, challenge traditional identity-based security models by impersonating humans or requiring user credentials, creating vulnerabilities in digital systems.