Google DeepMind AlphaProof Code Formally Verifies Signal Protocol Implementation
Google DeepMind introduced AlphaProof Code, a multi-LLM agentic framework that combines human expertise with AI agents to formally verify complex software. In its first major demonstration, the framework produced machine-checked Lean proofs that Signal's core cryptographic code satisfies key security properties — the first time AI has been used to formally verify a widely deployed cryptographic protocol at this depth, and the first time such a protocol implementation has been formally verified by any method, with or without AI.
What AlphaProof Code is
AlphaProof Code is a novel agentic framework for formal verification at unprecedented depth. It is designed to combine human expertise with AI agents rather than replace human judgment.
The framework splits formal verification the way a research team would: human experts work with AI to decide what to prove, and a group of AI agents carries out the proofs. In the Signal project, researchers collaborated with a mix of models to translate Signal's security goals into formal mathematical specifications, with human experts reviewing every specification. Agents powered by Gemini Carbon then generated the proofs.
How the Signal verification worked
The Signal verification followed a three-part workflow: specification, code, and proof. Researchers first translated Signal's security goals into Lean specifications, with human review of every one. The agents then proved that the compiled code fulfills those specifications, down to the exact code running on devices.
The proof effort generated over six million lines of Lean, a computer language that mathematically proves the accuracy of code. Lean automatically checks the generated proofs. The agents also audited the proof and highlighted areas needing a second look by a human — just one percent of the codebase. That allowed experts to focus review where human judgment mattered most.
Most protocol verification stops at the design level. AlphaProof Code went further by proving that the compiled implementation meets the key security properties, such as keeping messages protected before and after a device is temporarily compromised. Real-world vulnerabilities often hide in the gap between design and implementation, and bridging that gap makes the proofs vastly more complex.
Even at that complexity, the verification process took just four weeks. Google DeepMind says such a task historically takes expert teams years.
Why the Signal result matters
Cryptographic protocols underpin nearly all digital communication, from online banking to web browsing. A single flaw can expose millions of people at once, which is why experts have spent decades formally analyzing them. Signal's protocol protects billions of private conversations in apps such as Signal and Google Messages.
Signal CTO Ehren Kret described the result as a major step for practical assurance:
While prior formal verification focused on smaller portions of the code, Google DeepMind's approach enables significant expansion. 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.
Google DeepMind positions the work as a way to bring elite security assurance to businesses, governments, and critical infrastructure, and to help defend against agent-driven cyber threats. The company says verifying Signal's core cryptography lays groundwork for securing other vital systems, from medical device firmware to power grid software.
Testing whether the system catches real bugs
To confirm that AlphaProof Code can distinguish correct code from buggy code, Google DeepMind reintroduced two real, previously fixed bugs into an isolated version of Signal's code. In both cases, the system produced a machine-checked proof that the buggy version was insecure.
That validation matters because formal verification is only useful if it can detect failures as well as confirm successes. The Signal result therefore rests on two claims: the verified code satisfies the specified security properties, and the framework can produce machine-checked evidence when a version violates them.
Open artifacts and next steps
Google DeepMind is open-sourcing its formal verification artifacts for community use. The technical paper is titled "Formal Verification of Signal 4" and is available from Google DeepMind.
The company says scaling these techniques requires continued research. Paired with automated defense agents such as CodeMender, formal verification can help counter the speed and scale of modern cyber threats by creating software with mathematical proof that its protections work.
What AlphaProof Code does not claim
The announcement does not claim that AlphaProof Code replaces human verification experts or that it can verify arbitrary software without human involvement. Human experts reviewed every formal specification, and agents flagged one percent of the codebase for human review. Lean automatically checked the generated proofs.
The result is specific to Signal's core cryptographic code and the security properties specified by the researchers and Signal. Google DeepMind presents it as a milestone and a foundation for broader verification work, not as a general-purpose guarantee for all software.