We don’t imprison humans preemptively based on the capability to commit crime. Why regulate AI that way?

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

Here's a better framework. ...