Anthropic has put forward a proposed AI safety bill that would only impose its strictest requirements once an AI system has caused 50 deaths or $1 billion in damages, according to a report from Tom's Hardware. The proposal immediately raised a pointed question in the headline itself: would that be enough to keep an AI model from going out of control?
The idea is a liability-style trigger: below the threshold, AI developers would face lighter obligations; above it, tougher safety rules would apply. Anthropic has positioned itself as one of the most safety-focused AI labs, so a bill carrying such a high bar for action is already being read as a signal of where the company thinks regulation should land.
What the Anthropic AI safety bill would do
As reported by Tom's Hardware, the central mechanism of Anthropic's proposal is a damage threshold. The bill's heavier safety duties would only activate once an AI model is linked to 50 deaths or $1 billion in damages, framing regulation as a response to proven harm rather than a precaution taken in advance.
That structure puts the proposal in the middle of a live argument. Supporters of threshold-based rules say they avoid strangling early-stage research with compliance costs. Critics counter that waiting for 50 deaths before acting is, by design, waiting for a catastrophe that safety rules are supposed to prevent.
Why the 50 deaths or $1B threshold is controversial
The numbers themselves are the story. A 50-death trigger is far higher than the casualty counts that typically prompt new rules for cars, drugs or industrial equipment, and $1 billion in damages is a figure most companies would struggle to survive, let alone treat as a warning line.
The proposal also arrives as AI regulation is being debated in the United States, the European Union and elsewhere, with governments weighing how hard to push on model evaluation, transparency and liability. Anthropic's draft gives those debates a concrete number to argue over, which is likely why it spread quickly.
