I'm pretty skeptical of "AI safety research" being a useful field of inquiry. I'm guessing the the problems AIs create among themselves will be similar to the ones Humans create. Things may start changing so fast it gets disorienting though.