Anthropic is backing AI safety legislation in which the intervention threshold is set at at least 50 deaths or $1 billion in damages. The figure has drawn heavy coverage, much of it hostile, with headlines framing it as a bill that "would allow 50 deaths before AI model shutdown".
Both readings are available and it is worth separating them. Statutory thresholds exist to define when extraordinary powers engage, and every safety regime has one - the alternative to a number is a regulator with unbounded discretion, which industry opposes and legislatures rarely grant. Read that way, 50 deaths or $1 billion is a definition of catastrophic harm, not a permission slip.
Read the other way, a company writing its own trigger point into proposed law has picked a number below which nothing happens, and it has picked it in a domain where the harms are novel and the causal chains are contested. Both things are true at once, which is why the coverage has split so sharply. What the reporting does not settle is whether the threshold governs shutdown powers specifically or a broader set of interventions - and that distinction changes the argument entirely.
Sources
Yahoo News · Truthout