Google DeepMind verifies Signal’s core cryptographic code with AI
Google DeepMind says its researchers used AI agents to mathematically verify key security properties in the code behind the Signal protocol. The company describes this as the first formal verification of a widely deployed cryptographic protocol at this depth, and the first formal verification of such an implementation by any method. Signal technology protects private conversations in apps including Signal and Google Messages.
The work used AlphaProof Code, a multi-LLM framework in which human experts define and review formal security specifications while AI agents produce the proofs. Agents powered by Gemini Carbon generated more than six million lines of Lean, whose automated checks validate the proofs. The system also identified sections requiring additional human review, amounting to about one percent of the codebase.
The verification covered compiled code and properties including protection of messages after a temporary device compromise. It took four weeks, compared with years for similar expert-led work, according to the article. Researchers reintroduced two previously fixed bugs into an isolated Signal version, and the system proved both buggy versions insecure. Signal CTO Ehren Kret called the translation of live code into Lean theorems a major advance. Google DeepMind said it is open-sourcing the verification artifacts.
Written by Sakhanda Wire AI from reporting by DeepMind. Facts, quotes and full context are in the original article below.