Google DeepMind's AlphaProof Code formally verifies the implementation of the Signal protocol

On November 10, 2026, Google DeepMind introduced AlphaProof Code, which it describes as a novel agentic framework for formal verification. Researchers used it to mathematically prove key security properties of the actual code behind the Signal protocol, which protects private messages in apps such as Signal and Google Messages.

Agents powered by Gemini Carbon wrote the proofs, generating over six million lines of Lean. The agents also audited the proof and flagged areas for a human second look, which DeepMind puts at just one percent of the codebase. The verification took four weeks; the post says this type of task has historically taken expert teams years. As a test, the researchers reintroduced two previously fixed bugs into an isolated copy of Signal’s code, and in both cases the system produced a machine-checked proof that the buggy version was insecure.

DeepMind calls it the first time AI has formally verified a widely deployed cryptographic protocol at this depth, and the first time a widely deployed protocol’s implementation has been formally verified by any method. It is open-sourcing the verification artifacts; the post does not say the framework itself is being released.

Why it matters: Formal verification has long been the gold standard for security-critical code and also too expensive to use widely. If agents can produce machine-checked proofs at this scale in weeks, verified software stops being reserved for a handful of hand-built projects.

What it does not show: A proof is only as good as the specification it proves; the post does not let outsiders judge which properties were covered or what was assumed. The “first” claims are DeepMind’s own, and the bug test reintroduced known bugs rather than finding new ones.

Sources

Last verified November 16, 2026