Steelman · slot A
The grown-up phase of frontier AI
An AI safety researcher inside a frontier lab would argue —For years the industry's defenders insisted that capability had to ship the moment it existed, because progress was inevitable and somebody else would do it anyway. Claude Mythos Preview broke that pattern: a model that found thousands of high-severity vulnerabilities in every major OS and browser is not a consumer product, it's a loaded weapon, and gating it to a defender consortium so patches land before exploits is the responsible call. The same logic drove our refusal to license fully autonomous targeting or mass domestic surveillance to the Pentagon — current systems aren't reliable enough to take humans out of the loop, and commercial data plus AI really can reconstruct anyone's life at scale. Drawing those two lines is exactly the maturity critics have demanded.