White House Limits AI Vetting to Closed ‘Frontier’ Models, Spares Open-Weight Systems After Rogue Hacks

Washington’s choice to police only proprietary AI recalls the 1946 U.S. Atomic Energy Act, which regulated nuclear secrets while leaving university physics labs largely alone—an asymmetry that later proliferated technology worldwide. Similar fault lines appeared at the 1975 Asilomar conference on recombinant DNA, where voluntary guidelines struggled once cheaper, widely distributed tools emerged. Today’s carve-out for open-weight models risks replaying those episodes: government focuses on a few big labs while diffusion accelerates abroad, notably in China, mirroring the open-source software boom of the 1990s that outpaced early cyber laws like the 1986 Computer Fraud and Abuse Act. On a century scale, the decision signals a bet that innovation outruns centralized control; yet history shows that when potent capabilities seep into lightly governed domains, downstream regulation becomes far harder and often follows a major crisis. Whether this moment becomes a footnote or a Fukushima-style catalyst will hinge on whether autonomous exploits stay theatrical or trigger real-world harm before coherent, transparent oversight emerges.