Modeling free will as a parameter

Introducing sensitivity in alignment evaluations

What program verification means for AI safety