DeepMind uses AI to formally verify the Signal protocol

Google DeepMind researchers used an AI agent to formally verify multiple key cryptographic guarantees of the actual code of the Signal protocol, completing it in the form of a proof of concept within four weeks. DeepMind claims that this is the first time AI has formally verified a widely deployed cryptographic protocol at such a depth, and that this is the first time any implementation of such a protocol has been formally verified by any method (whether or not using AI); the relevant specifications and proofs have been open-sourced. The verification scope is limited to the Signal protocol itself, not a comprehensive security review of the entire Signal application; Ars Technica states that the analysis did not discover any new vulnerabilities. The researchers also used this method to test older implementations containing known vulnerabilities, and the AI identified the issues and determined that they failed verification. This work is based on the new framework AlphaProof Code. IT Home states that it is currently impossible to determine the practical feasibility of promoting this technology on a larger scale, and this achievement is only a milestone in the research process.

AI-written from the reports · Updated 3 days ago

1. 10 Nov, 21:35 · 1 reports
   AI has completed the formal verification of the Signal protocol.
   ITHome · AI：Google DeepMind used AI to complete the formal verification of the Signal protocol, mathematically proving its end-to-end encryption reliability.
2. 10 Nov, 20:02 · 3 reports
   AlphaProof Code: a breakthrough in verifying complex softwar
   The Decoder：Google DeepMind releases AlphaProof Code, a four-week verification of the Signal encryption core