AI-assisted proof gives Signal's encryption design a formal security check
Google DeepMind researchers have completed a formal verification of Signal's messaging security. They claim it is the first time AI has helped produce a proof that links protocol-level security guarantees to a widely used cryptographic implementation. The work used AlphaProof Code, a multi-agent framework in which a human-guided agent writes a specification in the Lean proof language, assesses machine code behaviour, and produces a proof that the Lean kernel checks against that specification.
The team verified Signal's post-quantum PQXDH handshake and its Triple Ratchet, which supports validation of libsignal's forward secrecy, authentication, post-compromise security and handshake secrecy. Humans reviewed the mathematical specification of Signal's security goals. Gemini Carbon agents then generated 6.5 million lines of Lean code to prove the libsignal implementation correct. An auditing pass flagged parts that needed human review. The work is a step toward a challenge the Beneficial AI Foundation issued in April to formally verify the Signal protocol and app with Lean.
There are limits. The result is theoretical because it assumes perfect cryptography and does not guarantee the computational secrecy of bytes. Even so, The Register describes it as a meaningful assertion of correctness. Researcher Swarat Chaudhuri said the system is built for human and AI collaboration, and that the team had to work out what to prove, rather than being handed a well-defined problem.