Formal Verification of Signal’s Protocol with Gemini

Why it matters

Google DeepMind researchers have used AlphaProof Code, a multi-LLM agentic framework, to formally verify the security properties of Signal's cryptographic protocol code — a task long viewed as a grand challenge in software security. The work is notable both as the first time AI has formally verified a widely deployed cryptographic protocol at this depth, and as the first time such a protocol's implementation has been formally verified by any method. Human experts worked with AI models to translate Signal's security goals into formal mathematical specifications, while Gemini-powered agents generated Lean proofs covering the codebase. Unlike most prior verification, which stops at the design level, the agents proved that compiled code running on devices satisfies key security properties, closing the gap where real-world vulnerabilities often hide. The system also detected deliberately reintroduced bugs, producing machine-checked proofs of insecurity. DeepMind is open-sourcing the verification artifacts and suggests the approach could extend to medical devices, power grids, and other critical systems.