The Mythos situation is what a responsible scaling policy gap looks like in practice

Anthropic built a model that found thousands of OS and browser vulnerabilities. Then they restricted it to roughly 50 critical-infrastructure organizations via Project Glasswing. Then Pete Hegseth designated Anthropic a supply-chain risk after the company refused military uses of Claude. Now the White House is weighing pre-release government vetting, possibly with the Pentagon running safety testing, and a 16-page executive order that would prohibit the private sector from interfering with federal AI use.

This is the sequence. A lab builds something genuinely dangerous, makes a unilateral call about who gets access, and the executive branch responds by improvising a framework in real time. The discussions reportedly included Anthropic, Google, and OpenAI executives, and the proposed approach may borrow from a U.K. model that assigns specific government bodies to evaluate whether models meet safety standards. Microsoft, xAI, and Google DeepMind have already made voluntary handovers for government review. Nobody had to do that under any binding rule. They did it because the alternative was worse.

Responsible scaling policies were supposed to be the industry's answer to exactly this. They were written by labs, reviewed by labs, and enforced by labs. Mythos made the gap visible to people who can sign executive orders. The oversight now being proposed is real. It is arriving because one model scared the right people at the right moment, not because anyone built a durable framework that could have handled the next one.