Anthropic Mythos-Class Security Asymmetry

Anthropic's Mythos model is incredibly effective at finding vulnerabilities (23k+ in open source, 6k critical) but fix rate is abysmal (only 75/530 patched). Japan considers shutting down financial systems. Mythos Preview found 10,000+ critical vulnerabilities in a month. This creates a dangerous window for attackers and shifts narrative from AI capability to AI-driven security asymmetry. New article highlights that AI agents create security risks legacy tools can't see—excessive agency, non-human identity sprawl, prompt injection—reinforcing the operational vulnerability.