Anthropic pitches AI safety bill with court check

Anthropic has pitched an AI safety bill that could block the release of a model if more than 50 people could be killed, according to TechCrunch. The proposal is presented as a way to protect against dangerous AI systems, but the report says it would also prevent federal agencies from blocking a model’s release without a judge’s permission. That structure is the key tension in the story. On one side, the bill would create a high-stakes safety threshold and give regulators a tool to intervene before a model reaches the public. On the other, requiring judicial approval could slow emergency action and add legal constraints at exactly the moment when speed might matter. The proposal therefore captures a broader debate about how governments should regulate advanced AI. Supporters of court checks may argue that they prevent arbitrary agency decisions and ensure strong due process when a company’s product is at stake. Critics may argue that the threshold is difficult to define, that evidence about potential harm is often uncertain, and that waiting for a judge could make the safeguard too slow in a crisis. The 50-person threshold is also likely to draw scrutiny. Estimating how many people could be killed by a model is not a straightforward technical exercise, and the bill would need a process for making that judgment credible. The report does not provide the full legislative text, so details such as who makes the initial determination and what standard a court would apply remain unclear. Still, the story is significant because it shows a leading AI safety company engaging directly with policy design. It also illustrates the challenge of writing rules for systems whose risks are hard to measure in advance. As AI regulation debates continue, the Anthropic-backed proposal is likely to be cited both as a serious safety effort and as a caution about legal process and emergency response.