Anthropic pushes for law letting courts stop imminent AI releases
Anthropic, a big name in AI research, is pushing for a new law that would allow courts to stop release only under specific imminent-threat criteria, if an assessment finds that a model poses a foreseeable and imminent threat of more than 50 potential deaths or property damage exceeding $1 billion.
Their goal? Make sure dangerous tech doesn't get out into the world unchecked.
Commerce-registered evaluators alert within 72 hours
The bill lays out a process where independent evaluators (registered with the US Commerce Department) check if an AI model could be seriously harmful, like enabling cyberattacks or even helping create biological weapons.
If they spot a real threat, they have 72 hours to alert authorities, who can then ask a court to stop the model from being released.
Anthropic hopes this approach will keep AI safer while letting innovation continue.