AI alignment researchers estimate less than a 10% chance that we lose control of superintelligent AI once it’s built. More typical estimates range from 10-80%, depending on who you ask.