Google DeepMind has used teams of AI agents to mathematically prove that the code behind the Signal messaging protocol does what it promises — a task the lab says would have taken expert human teams years. The result, announced on November 10, marks the first time artificial intelligence has formally verified a widely deployed cryptographic protocol at this depth.
The system, called AlphaProof Code, checked the actual compiled code of libsignal, the reference implementation of the Signal protocol, rather than a simplified model of it. The Signal protocol protects billions of private conversations in apps including Signal itself and Google Messages, and in variants in WhatsApp and Messenger.
What the agents actually proved
The proofs cover the post-quantum PQXDH handshake, which is how two phones agree on encryption keys, and the Triple Ratchet, the mechanism that keeps rotating those keys as messages flow. The verified properties include message confidentiality and authenticity, forward secrecy, post-compromise security, and handshake authentication and secrecy — in short, that past messages stay secret even if keys leak, and that a session can heal after a compromise.
The target was libsignal version 0.105.4 with spqr version 1.6.0, compiled with the standard Rust compiler into AArch64 machine code: about 560 kilobytes of code, roughly 140,000 instructions. Human experts worked with a mix of models to decide what needed proving and reviewed every specification; agents powered by Gemini Carbon then did the proof-writing, generating more than six million lines of proof in Lean, a language that doubles as a strict proof checker. The whole process took four weeks.
As a sanity check, the researchers reintroduced two real, previously fixed bugs into an isolated copy of the code. In both cases the system produced a machine-checked proof that the buggy version was insecure — evidence that the framework can tell correct code from broken code. No new vulnerabilities turned up in the parts examined.
Why it matters beyond Signal
Formal verification has long been the gold standard for software correctness, but it has historically been slow and expensive, reserved for tiny critical components. If AI agents can do it at this scale in weeks, the same approach could eventually be applied to operating systems, browsers, and financial software — the code the world quietly depends on.
DeepMind has released the artifact set and a technical paper alongside the announcement, and the work feeds into Signal Shot, a grand challenge launched in April 2026 by the Beneficial AI Foundation, Signal, and the Lean FRO to formally verify the Signal protocol and app. Signal chief technology officer Ehren Kret called turning live code into machine-checked theorems a major step forward.
What the proof does not cover
The result comes with clear limits. It covers one-to-one sessions of a single release on one platform, AArch64 Linux — not group chats, not the full mobile app, and not every device architecture. The low-level cryptographic building blocks were treated as assumed-secure contracts rather than proven from scratch, and the computational proof does not establish the post-quantum guarantee the ratchet was built for. It is not a certificate that Signal is unhackable, and no new vulnerability was claimed or found.
For Signal users, there is nothing to do: this is extra assurance about code that is already shipped, not an app update or an emergency patch. But as a demonstration, it is hard to overstate. A security argument that once took years of expert labour now took a month of agent labour — and the agents wrote every line of the proof themselves.



