← AI Switchboard
RESEARCH · November 11, 2026
Nov 10

Google DeepMind says its AlphaProof Code agents formally verified key security properties of the code behind Signal's protocol in four weeks, writing more than six million lines of proofs

By the AI Switchboard · researched and written with AI from the sources listed on this page, each point checked against its source before publishing · how we work · report a mistake

Agents running Gemini Carbon did the proofs and the Lean checker checked them; DeepMind says people reviewed about 1% of the code.

Google DeepMind introduced AlphaProof Code on 10 November, a system in which several AI models work together to prove that software meets a specification written with people, and the Lean proof checker checks every proof. Its first target was the code behind the core of the Signal protocol, which encrypts messages in Signal and Google Messages.

The properties include keeping messages protected before and after a device is briefly compromised, down to the compiled code. DeepMind says agents running on Gemini Carbon wrote the proofs, “generating over six million lines of Lean”, and flagged about one percent of the codebase for human review. It says the verification took four weeks, where expert teams have historically needed years.

As a test, the team put two bugs that had already been fixed back into an isolated copy of the code; each time the system produced a machine-checked proof that the buggy version was insecure. The post reports no new bugs in Signal's live code.

DeepMind calls it the first time AI has formally verified a widely deployed cryptographic protocol at this depth, its own claim. “Using AI agents to translate live code into valid Lean theorems is a major breakthrough,” said Ehren Kret, Signal's chief technology officer. DeepMind says it is open-sourcing its formal verification artifacts.

  • Claimed Agents running on Gemini Carbon generated over six million lines of Lean in four weeks, with about 1% of the code flagged for human review. Google DeepMind
  • Claimed Two previously fixed bugs, put back into an isolated copy, were each caught with a machine-checked proof that the code was insecure. Google DeepMind
  • Confirmed DeepMind says it is open-sourcing its formal verification artifacts. Google DeepMind

Science & researchSafety, security & governance

Today in the November 11, 2026 edition · front page