[Question] Is anyone working on formally verified AI toolchains?

Not talking about solving alignment, but preventing stuff like “we solved alignment but we died anyways because of a race condition.”