Anthropic's draft safety bill

News · Heat 1.4 / 10 · 2 shows · 12m airtime

Politico reported on October 19, 2026 that Anthropic is circulating a confidential draft AI safety bill on Capitol Hill. The draft would let the government block a model release posing foreseeable imminent risk of more than $1 billion in damages or 50 deaths from a bioweapon, loss of developer control, or a cyberattack, with blocking adjudicated in federal court. It also bars any other agency from suspending a model's development or use and sets the frontier at 10^25 flops. Safety advocates and critics have attacked its definitions and its preemption of state rules.

How 2 podcasts covered it

1. The AI Daily Brief, 2026-10-19 (26:41 to 31:01)
   Walks through the leaked safety bill draft
   Walks through the leaked draft: the $1 billion and 50-death thresholds, evaluators registered with Commerce, federal-court adjudication, and language barring other agencies from suspending a model. Collects the pushback from safety advocates and from David Sachs on the 10^25 flops frontier definition.
   New angle: The draft bars any other agency from suspending a model's development or use, and sets the frontier at 10^25 flops.
2. What Next, 2026-10-23 (8:01 to 13:48)
   Walks through the leaked draft and the confusion
   Walks through the leaked Anthropic draft and the confusion it sows on Capitol Hill: lawmakers hear from AI safety groups and AI companies without realizing the same funders back both. Argues the superintelligence framing is 1990s futurism, and reads the leak's timing as image control before the midterms.
3. What Next, 2026-10-23 (13:48 to 15:11)
   Weighs preemption inside the draft debate
   Weighs the preemption question inside the Anthropic draft debate, recalling a past federal preemption push that Marsha Blackburn helped knock down after music companies objected. Newport's warning about industry using preemption to freeze out stronger state rules is argument, not a new development.