Coding is Love
News 2 min read

DeepMind's AlphaProof Code formally verifies Signal's protocol code

Google DeepMind has announced AlphaProof Code, an agentic framework that mathematically proved key security properties of the code behind the Signal protocol, the encryption used by Signal and Google Messages. The announcement came on November 10, 2026. DeepMind says it is the first time the implementation of a widely deployed protocol like this has been formally verified by any method, with or without AI, and that the work took four weeks instead of the years such efforts usually need.

What’s new

AlphaProof Code is a multi-LLM framework that splits formal verification the way a research team would. Humans and AI models together turn Signal’s security goals into formal specifications, and human experts reviewed every one. Agents powered by Gemini Carbon then did the proofs, generating over six million lines of Lean, a language whose proofs are checked automatically.

The agents also audited their own proofs and flagged the parts that needed a second human look: just one percent of the codebase, according to DeepMind. Lean checks the rest.

Most protocol verification stops at the design level. DeepMind says its agents went further and proved that the compiled code meets the specified properties, including keeping messages protected before and after a device is temporarily compromised.

To test that the system can tell good code from bad, the team reintroduced two real, previously fixed bugs into an isolated copy of Signal’s code. In both cases it produced a machine-checked proof that the buggy version was insecure.

Signal’s CTO Ehren Kret said that using AI agents to translate live code into valid Lean theorems is “a major breakthrough”. DeepMind is also open-sourcing its formal verification artifacts.

What it means for developers

You cannot install AlphaProof Code today: DeepMind describes this as research and says scaling the techniques needs continued research. What you can do is read the technical paper and the open-sourced artifacts to see how the specifications and Lean proofs are structured.

The practical takeaway is the workflow. Humans own the specification, agents write the proofs, and a proof checker rather than a person decides whether a proof is valid. That makes AI-written output verifiable, which is the opposite of trusting a model’s code because it looks right. DeepMind points to future targets such as medical device firmware and power grid software, and to pairing verification with its CodeMender security agent.

Two limits are worth keeping in mind. A proof is only as good as its specification, which is why humans reviewed every one. And the result covers Signal’s core cryptographic code and the properties DeepMind specified, not every line of every app that uses the protocol.

How it compares

DeepMind says earlier formal verification work focused on smaller portions of code and needed extensive manual effort. Its claim here is scale and speed: over six million lines of Lean and a four-week process. These figures come from DeepMind itself; the technical paper is the place to check the details.

For more on Google’s AI infrastructure this week, see our post on the TPU 8tx and TPU 8i Max.

Sources