THE CRUNCH
Google DeepMind researchers say they have formally verified the code behind Signal's messaging security, using a new multi-agent framework called AlphaProof Code. The team claims it is the first time AI has helped produce a proof linking protocol-level security guarantees to a widely used cryptographic implementation, and the first time any method has verified the implementation of a protocol deployed at this scale.
The framework splits the work like a research team would. Human experts collaborated with a mix of models to translate Signal's security goals into a mathematical specification written in Lean, a language that lets a computer check proofs automatically. Agents powered by Gemini Carbon then generated roughly 6.5 million lines of Lean code proving the libsignal implementation correct, before an auditing pass flagged the small portion needing human review. DeepMind puts that review burden at just one percent of the codebase.
The verification covered Signal's post-quantum PQXDH handshake and Triple Ratchet, validating libsignal's forward secrecy, authentication, post-compromise security and handshake secrecy. The result is theoretical in one respect: it assumes perfect cryptography and does not guarantee the computational secrecy of the bytes themselves. Even so, DeepMind says the process took four weeks with a team of five, against historical efforts like the seL4 microkernel verification that took 20 person-years.
As a sanity check, the team reintroduced two real, previously fixed bugs into an isolated copy of Signal's code, and the system produced machine-checked proofs that the buggy versions were insecure. Signal's CTO Ehren Kret said the approach enables a significant expansion beyond the smaller scopes prior formal verification had to accept. The verification artifacts are being open-sourced, and the work answers, in part, an April challenge from the Beneficial AI Foundation to formally verify the Signal protocol in Lean.