What we know

Anthropic has proposed that AI models should be blocked only if they pose a risk of causing more than $1 billion in damage or result in over 50 deaths or serious injuries. This threshold is part of a confidential draft obtained by Politico and reported by TechSpot. A company spokesperson described the proposal as containing "sensible policies" that Congress should consider.

Why it matters

Anthropic’s proposal sets clear, high-impact thresholds for when AI models might require regulatory intervention, focusing on preventing catastrophic harm. By suggesting that only AI systems with the potential for extremely severe consequences be blocked, the company aims to influence how lawmakers approach AI oversight.

What is still unknown

The proposal’s details on how these thresholds would be measured or enforced remain unclear. There is no information on the accuracy of risk assessments, which AI languages or models this would apply to, or when any related policies might be enacted. No independent evaluations of the proposal’s feasibility or impact have been reported.