Anthropic's proposed AI safety bill only kicks in if 50 deaths or $1B in damages

Commenters focused on Anthropic’s proposal to trigger government review only after an AI causes 50 deaths or $1B in damages, calling the thresholds arbitrary, easily gamed, and vague. Many mocked the idea by applying it to trolley problems and value-of-life math (some noting the implied ~$20M per life), while others raised concrete gaming scenarios - doubling currency value or splitting harms into many small events - and questioned whether the limits are cumulative, per-event, or time‑bounded. Several commenters worried about who would “keep score” and how “death” or damages would be defined, and some said the thresholds would turn safety certification into an expensive insurance/audit regime that favors large labs and blocks open-source projects.

Opinion split on remedies and intent: some urged criminal prosecution for deliberate AI-enabled harm and pointed to statutes like RICO or existing regulatory analogies (pharma, autos), while others argued unintended harms require regulation rather than punishment because mens rea is often absent. A number of commenters asserted the bill is regulatory capture - designed to protect big companies - and noted that determined actors can still run dangerous models on affordable hardware. Others defended audits and third‑party review as sensible in principle but warned about enforcement complexity and legal ambiguity over who should be liable (engineers, CEOs, or deployers).