Background
Formal verification uses machine-checkable mathematical proofs to show that software meets a precise specification. Writing such proofs is demanding, and the scale of the work determines how much of a codebase can be checked this way.[1]
What happened
On November 10, 2026, Google DeepMind introduced AlphaProof Code. Agents powered by Gemini Carbon generated over six million lines of Lean proof.[1]
The verification covered the compiled code of Signal's cryptographic protocol implementation, not only its design.[1] The verification took just four weeks.[1]
Key details
Human review covered about one percent of the codebase, namely the areas the agents flagged.[1]
To test whether the system could detect real problems, DeepMind reintroduced two previously fixed bugs into an isolated version of Signal's code. The system produced machine-checked proofs that both versions were insecure.[1]
DeepMind is open-sourcing its formal verification artifacts for community use. The announcement does not introduce a product or API.[1]
Impact
The project shows agents producing and checking formal proofs for a full cryptographic codebase in a short time. Because the verification artifacts are open-sourced, other researchers and developers can examine and reuse the work. The announcement describes no product or API, so the artifacts themselves are the main outcome.[1]
Sources
- 1
Cite this page: Google DeepMind introduces AlphaProof Code for formal verification of compiled software. TellWorld, https://www.tellworld.net/news/alphaproof-code-2026. Updated 2026-11-15