Google DeepMind says a new AI system, AlphaProof Code, produced machine-checked proofs that key security properties of the Signal messaging protocol hold in its code. These are DeepMind's claims, published on its own blog. We found no independent verification yet, so read every result here as DeepMind's account.

What DeepMind says it did

  • Method: AlphaProof Code is described as a multi-LLM agentic framework. Agents powered by what DeepMind calls Gemini Carbon (the model in our Gemini 4 Carbon story, if it is the same one; DeepMind does not say) wrote proofs in Lean, a language that a computer can check line by line.
  • Scale: DeepMind says the agents generated over six million lines of Lean and flagged about 1% of the codebase for human review.
  • Specifications: it says human experts reviewed every formal specification, meaning the written statements of what the code must do.
  • Claim: DeepMind says it proved the compiled code meets key security properties, such as protecting messages before and after a temporary device compromise, "down to the exact code running on devices".
  • Time: about four weeks.
  • Bug test: DeepMind says it put two real, previously fixed bugs back into an isolated copy and the system produced machine-checked proofs that the buggy version was insecure.
  • Release: DeepMind says it is open-sourcing the formal verification artifacts. Its artifacts page is linked from the post.

What Signal and others say

Signal's chief technology officer, Ehren Kret, is quoted on DeepMind's page praising the approach. In a statement to Wired, he said that verifying the Signal protocol implementation "gives us even greater confidence in our code's privacy guarantees." We found no post on signal.org/blog about the work.

Wired reports the analysis was limited in scope: it did not formally verify the entire Signal mobile app, and DeepMind researcher Swarat Chaudhuri told Wired there is "a long distance to go" before this scales to all the software of the world. Wired also reports it found no new vulnerabilities in the portions examined, and that earlier Signal Foundation verification work gave the team a head start. Google declined to share the cost or say whether a commercial product is planned.

What is not known

  • Independent checks: we have not seen outside researchers confirm the proofs.
  • Scope: the proofs cover the code and properties DeepMind names, not Signal as a whole, the app, or every possible attack.
  • Cost and practicality: unknown.

What this means for you

If you use Signal, nothing changes in the app today and you need not do anything. This is a research result about checking code, not a new feature or a security warning. For account protection you control today, see our passkeys guide. We draw no conclusion about whether the proofs will hold up under outside review.

What we checked

On November 10 we read DeepMind's post and the Wired report, and searched Signal's blog for a statement. We did not read the technical paper, run the proofs or check the Lean code.