On 10 November 2026, Google DeepMind introduced AlphaProof Code on its official website. AlphaProof Code is a formal-verification framework that combines human experts with AI agents. It is not a new chat-model launch.
Researchers used the framework to prove selected key security properties of libsignal, the Signal protocol’s reference implementation, and carried those proofs through to compiled machine code / AArch64. The official blog describes the technology as protecting private conversations in apps like Signal and Google Messages. The paper says the results concern this build of libsignal-cli, not Signal’s own applications.
The technical paper and the artifacts page say the agents generated about 6.5 million lines of machine-checked Lean proof code. The official blog body says the agents generated over six million lines of Lean.
The paper says the proofs cover Signal’s post-quantum PQXDH handshake and Triple Ratchet, establishing forward secrecy, authentication, post-compromise security, and handshake secrecy. It describes the results as the first proofs connecting protocol-level security guarantees to the unmodified code of a widely used cryptographic protocol implementation.
The proofs were produced by agents powered by Gemini Carbon (named Gemini 4 Carbon in the paper). Researchers collaborated with a mix of models to draft formal specifications, and human experts reviewed every one. The paper also names Claude Opus 5.5 in the specification stage, and says that none of the authors wrote any proof code; the proofs were produced by Gemini 4 Carbon agent teams. For the existing introduction of the model used in these proofs, published on 28 October 2026 — not a release today — see Google Announces Gemini 4 Carbon, Next Frontier Model. This introduction of AlphaProof Code is not a new Carbon API or pricing release.
The official blog says the verification process took just four weeks. The paper adds that the proof work was done by a team of five over four weeks.
Google DeepMind says this is the first time AI has been used to formally verify a widely deployed cryptographic protocol at this depth, and the first time the implementation of a widely deployed protocol like this has been formally verified by any method, with or without AI. That is not a claim that the entire Signal application, or every possible failure mode, has been verified.
To check that the framework can tell correct code from buggy code, the researchers reintroduced two real, previously fixed bugs into an isolated version of the code. In both cases, the system produced a machine-checked proof that the buggy version was insecure. That is a regression check, not a report of new Signal vulnerabilities.
“While prior formal verification focused on smaller portions of the code, Google DeepMind's approach enables significant expansion,” said Ehren Kret, Chief Technology Officer of Signal. “Using AI agents to translate live code into valid Lean theorems is a major breakthrough. As AI-driven security threats grow, rigorously verifying the Signal protocol implementation gives us even greater confidence in our code's privacy guarantees.”
The formal verification artifacts and the technical paper have been released.