Can I download the AlphaProof Code verification files?

DeepMind says it is open-sourcing the formal verification artifacts for the Signal work. They are published on a Google DeepMind GitHub Pages site, with a technical paper as a PDF. The announcement does not mention an API or a commercial product for AlphaProof Code itself.

How much of the Signal code did humans have to review?

AlphaProof Code left human reviewers with about one percent of the Signal codebase. The agents audited their own proofs and flagged those areas for review, while Lean automatically checked everything else. Human experts also reviewed every formal specification that defined what the code must do.

Did AlphaProof Code find real bugs in Signal?

The announcement does not report new Signal bugs. Instead, the team re-introduced two previously fixed bugs into an isolated copy of Signal's code. For both, AlphaProof Code produced a machine-checked proof showing the buggy version was insecure.

Does the proof cover the real code running on phones?

DeepMind says the AlphaProof Code proofs go beyond protocol design and cover the compiled code that runs on devices. One example in the post is that messages stay protected before and after a device is temporarily compromised.

Google DeepMind · 2026-11-10 · major

AlphaProof Code — DeepMind AI verifies the Signal protocol's code

In short

AlphaProof Code is a DeepMind agent framework that writes Lean proofs that code is correct. It verified the Signal protocol's implementation in four weeks, and DeepMind is open-sourcing the verification artifacts.

Google DeepMind · 2026-11-10 · 2 sources checked

Illustration from the Google DeepMind AlphaProof Code announcement — Google DeepMind

A multi-LLM agent system that produces machine-checked Lean proofs that real software is correct.

Quick facts

Maker	Google DeepMind
Proof language	Lean
Proof agents	Gemini Carbon
Target verified	Signal protocol implementation
Time taken	4 weeks
Output	6M+ lines of Lean
Availability	Verification artifacts open-sourced

What is it?

AlphaProof Code is a multi-LLM agentic framework from Google DeepMind for formal verification, which means mathematically proving that code does what its specification says. DeepMind used it on the code behind the Signal protocol, the encryption that protects messages in Signal and Google Messages. The post calls this the first time AI has formally verified a widely deployed cryptographic protocol at this depth.

How does it work?

Human researchers and a mix of models first turn Signal's security goals into formal specifications, and human experts review every one. Agents powered by Gemini Carbon then write the proofs in Lean, audit their own work and flag parts for human review. Lean checks the proofs automatically, and they cover the compiled code that runs on devices, not only the protocol design.

Why does it matter?

Verifying code this deep normally takes expert teams years, and DeepMind reports four weeks. The system generated over six million lines of Lean, while humans reviewed only about one percent of the codebase. In a test, it also produced machine-checked proofs that two re-introduced bugs made Signal's code insecure.

Who is it for?

security engineers, cryptography and formal-methods researchers

Frequently asked questions

Can I download the AlphaProof Code verification files?: DeepMind says it is open-sourcing the formal verification artifacts for the Signal work. They are published on a Google DeepMind GitHub Pages site, with a technical paper as a PDF. The announcement does not mention an API or a commercial product for AlphaProof Code itself.
How much of the Signal code did humans have to review?: AlphaProof Code left human reviewers with about one percent of the Signal codebase. The agents audited their own proofs and flagged those areas for review, while Lean automatically checked everything else. Human experts also reviewed every formal specification that defined what the code must do.
Did AlphaProof Code find real bugs in Signal?: The announcement does not report new Signal bugs. Instead, the team re-introduced two previously fixed bugs into an isolated copy of Signal's code. For both, AlphaProof Code produced a machine-checked proof showing the buggy version was insecure.
Does the proof cover the real code running on phones?: DeepMind says the AlphaProof Code proofs go beyond protocol design and cover the compiled code that runs on devices. One example in the post is that messages stay protected before and after a device is temporarily compromised.

Try it

https://google-deepmind.github.io/code-verification