White House weighs pre-release AI checks

Mythos is less a product launch than a policy trigger: Anthropic is effectively demonstrating that frontier models can cross from “helpful coding tool” into “dual-use cyber weapon,” which forces regulators to think about gatekeeping before deployment.

• –Anthropic says Mythos Preview found thousands of zero-day vulnerabilities and is not planned for general availability.
• –The model is being used in Project Glasswing with selected partners to harden software and critical infrastructure.
• –Axios reports the White House is discussing pre-release safety testing for models deployed into public-sector environments.
• –This is a strong sign that cyber capability is becoming a first-class AI governance issue, not just a lab-safety footnote.