Dario Amodei is right that safeguards need time to catch up with AI capabilities. But companies deploying agents face a related question: how much authority should they hand over?
When model output becomes real action, AI safety doesn’t stop at the model.
