Proving safety for narrow AI outputs

Identifies domains where AI can deliver capabilities with quantitative guarantees against objective safety criteria. Maps a path to generating software with formal proofs of specification compliance.