Axiom’s zero-knowledge proofs might one day help detect deepfakes