Breaking01 / AI
Google DeepMind uses AI to verify Signal's core code

Google DeepMind said it verified Signal's core cryptographic code in four weeks against specified security conditions, in AlphaProof Code research published on 10 Nov.
The announcement described formal verification as mathematically proving how code behaves. Researchers said human experts reviewed the security specifications that the AI agents were tasked with proving.
According to Google DeepMind, agents powered by Gemini Carbon wrote the proofs. Lean, a computer language used for mathematical proofs, checks each proof to verify the agents' work.
The researchers said they reintroduced previously fixed bugs into an isolated version of Signal's code to test the system. It produced machine-checked proofs showing that the faulty code was insecure.
The announcement said most previous protocol verification stopped at the design stage. This work extended verification to compiled code—the code that runs on devices—to prove that it met the specified security conditions.
Google DeepMind did not disclose a specific timetable for extending verification to other systems.