one way to address AI risk might be very strict liability early on (already arguably we are late to the game) for whomever deploys it.