A Defendant, Not a Pause
Near-term AI alignment is a defendant. Whoever deploys an agent that causes harm is liable for that harm. A government pause is what you ask for when the subscription is threatened, the open weights are cheaper, and the next training run …
