AlphaProof Code (deepmind.google)

Google DeepMind researchers have achieved a significant milestone in software security by using their new framework, AlphaProof Code, to mathematically verify the cryptographic protocol underlying Signal, which secures billions of private messages. This breakthrough represents the first time AI has been employed to conduct formal verification on a widely deployed cryptographic protocol at such an unprecedented depth, successfully proving that the compiled code upholds critical security properties that protect communications from potential vulnerabilities. AlphaProof Code operates through a collaborative approach, combining human expertise with AI agents to streamline the formal verification process. By breaking down the task, researchers first translated Signal’s security goals into formal mathematical specifications, which were rigorously reviewed. Subsequently, AI agents executed the proofs, generating six million lines of Lean code while identifying only one percent that required additional human review, significantly reducing the time needed from years to just four weeks. This technology not only strengthens the security assurance of digital communication but also highlights an innovative method to bridge the gap between protocol design and implementation, an area that has historically been a challenge for cryptographic security.