We don’t imprison humans preemptively based on the capability to commit crime. Why regulate AI that way? (opens in new tab)

Pre-deployment testing and capability assessments can't predict how AI systems actually behave in the wild. Here's a better framework.