Anthropic proposed an AI safety bill that would allow blocking a model if its deployment could lead to the death of 50 or more people. This according to TechCrunch AI; the available description is based on metadata, not the full text of the document.
According to the same description, federal agencies would not be able to halt the release of a model without a judge's permission. Thus, the proposed measure concerns not only the threshold of potential harm but also the procedure for government intervention.
The significance of the initiative depends on how the document defines risk, the link between the model and harm, and the powers of the court and agencies. These details are not disclosed in the provided source.
editorial commentary
Why it matters
A likely consequence is the transfer of some decisions about releasing high-risk models from the administrative realm to the judicial one. The next observable signal will be the publication of the full text of the initiative or an official government reaction. Significant uncertainty remains due to the absence of a full document and primary comments.