DeepMind AI Agents Prove Security Properties of Signal Protocol Code in Four Weeks

Google DeepMind announced on November 10, 2026 that it used AI agents to formally verify key security properties of the code behind the Signal protocol. According to the company, the proofs cover the compiled code that runs on devices, not just the protocol's design assumptions.

What was proven

The Signal protocol is used in the Signal app and in Google Messages, and according to 9to5Mac, WhatsApp uses it too. DeepMind says the agents proved, among other things, that messages stay protected before and after a device is temporarily compromised.

Formal verification means mathematically proving that a protocol meets its security claims. That is a higher bar than ordinary testing, which looks for bugs but cannot rule out that more exist.

How AlphaProof Code works

The system is called AlphaProof Code, a multi-agent framework built on multiple language models that splits the work between humans and AI. Researchers, working with a mix of models, translated Signal's security goals into formal specifications, and experts reviewed every one.

Agents powered by Gemini Carbon wrote the Lean proofs. The agents then audited the proofs and flagged the parts that needed a human look. According to DeepMind, Lean automatically checks every proof, which makes it possible to verify the agents' work.

Limits of the result

DeepMind says it verified Signal's cryptographic core and cautions that scaling these techniques requires further research. 9to5Mac stresses that the verification applies to the protocol, not the whole app. Human judgment is still needed for the specifications and the flagged parts of the proofs.

According to 9to5Mac, the process is very computationally demanding, and Google does not know how practical broader use will be. The resources used were not disclosed, so no one should assume other protocols will be verified soon. The company itself claims this is the first time a widely deployed cryptographic protocol has been formally verified at this depth with AI.

We are very excited by this result, but there is a long distance to go before we are able to scale this to all the software of the world, and we want to do that. - Swarat Chaudhuri, DeepMind research scientist, speaking to WIRED (as quoted by 9to5Mac)

Signal's reaction and next steps

Signal's chief technology officer (CTO) Ehren Kret said that using AI agents to translate live code into valid Lean theorems is a "major breakthrough", and that the verification gives "even greater confidence in our code's privacy guarantees". Rolfe Schmidt, Michael Rosenberg and Kret gave feedback on the methodology.

DeepMind is open-sourcing the verification artifacts for community use and says the approach could extend to other critical systems, such as medical device firmware or power grid software. For now, that is a statement of direction, not a deployment.