DeepMind uses AI to verify key parts of Signal’s encryption

Google DeepMind researchers used an agentic AI approach to mathematically verify key cryptographic guarantees of the Signal protocol in a four‑week proof‑of‑concept project. The work, performed with a new AI framework called AlphaProof Code, provides an independent validation of the cryptography protecting billions of communications on Signal, Google Messages and WhatsApp, and demonstrates AI‑assisted formal verification operating on a nontrivial portion of the protocol rather than the entire mobile app.

DeepMind emphasises that this is a milestone rather than a complete solution, and notes the analysis did not uncover new vulnerabilities in the examined components.

The researchers also tested older Signal versions containing known, long‑fixed vulnerabilities; in those cases the AlphaProof Code analysis correctly identified issues and showed the protocol could not be formally verified. While there was no cost disclosure for the project, the team released the specification and proofs to aid further formal verification work.

Signal’s chief technology officer, Ehren Kret, said the collaboration expanded prior verification work and could bolster confidence in the protocol’s privacy guarantees as AI‑driven security tools mature. Pushmeet Kohli, Google Cloud and DeepMind’s chief scientist, framed the work as illustrating a human–AI collaboration process that combines automated proof generation with essential human review, while acknowledging questions remain about scaling AI‑assisted verification for large software stacks.