Anthropic's proposed AI safety legislation requires a judge's approval before federal agencies can block a dangerous model's release.
Anthropic is pitching federal legislation framed as blocking AI models that could plausibly kill 50 or more people. The fine print does something different: it forbids federal agencies from halting a model's release without first getting a judge's sign-off, shifting enforcement power away from regulators and into the courts.
The framing matters because it reads as consumer protection while structurally weakening the fastest-acting check on model releases, agency authority, in favor of a slower, litigation-dependent process.
Model labs are increasingly writing the rules meant to govern them, and the details decide who actually holds the brake pedal. Leaders should watch this closely: a judge-gated process means releases move forward by default while agencies fight it out in court, favoring whoever ships first.
The daily signal, curated. Get it in your inbox.
Subscribe on LinkedIn →