Google DeepMind has introduced AlphaProof Code, an AI framework that formally verified the Signal protocol, generating over six million lines of Lean code with human experts reviewing only one percent.
Google DeepMind has introduced AlphaProof Code, an AI-powered framework for formally verifying complex software. The framework uses multiple large language models and AI agents working with human experts to mathematically prove that code operates correctly.
DeepMind used AlphaProof Code to formally verify the Signal protocol, an end-to-end encrypted communication protocol used in various messenger apps. The research team translated the protocol's security objectives into a formal mathematical specification, which was reviewed by human experts. An AI agent equipped with Gemini 4 Carbon then performed the proof, generating over six million lines of Lean code, a language used for mathematically proving code accuracy.
An AI agent examined the Lean proof and identified areas requiring human visual re-verification. These areas accounted for only one percent of the total, resulting in significant time savings. According to the source, a formal verification process that would normally require several years of work by experts was completed in just four weeks.
To test AlphaProof Code's ability to distinguish between correct and buggy code, the research team incorporated two already-fixed bugs into the code and had it re-evaluated. AlphaProof Code generated verification results indicating that the buggy version was unsafe.
Signal's CTO, Ehren Krett, said: "Traditional formal verification has inevitably been limited to a small scope, but this new approach allows for significant scaling. Previously, verifying that production code strictly matches the cryptographic design required a tremendous amount of manual work. Therefore, using an AI agent to automatically convert live code into valid Lean theorems is a major breakthrough. As AI-based security threats increase, rigorously verifying the core Signal protocol with automated theorem proving further increases our confidence that the code is reliably fulfilling privacy guarantees."
Pushmith Kohli, Vice President at DeepMind, said: "Beyond a research breakthrough, this is a prime example of how cutting-edge AI can be an amplifier to counter the speed and scale of modern cyber threats."
The research project focused on specific cryptographic guarantees and identified known vulnerabilities in older versions of the protocol. It did not encompass the entire Signal mobile application. Researchers emphasize that this is a milestone and not a complete security verdict for the entire app. Google DeepMind has published the formal specification and proof, allowing other researchers to build upon this work. Google has yet to disclose details on potential productization or service offerings.
Sources
This article was drafted with AI assistance and checked against the sources above. Company claims are reported as claims. Cover image is AI-generated.
The day's checked news, by email
One short email a day with the top stories. Free, and you can unsubscribe in one click.
Or follow for the latest news:
No comments yet. Start the discussion.
United Airlines will restrict unauthorized third-party AI agents from booking flights and managing reservations, citing cybersecurity risks and federal aviation rules.
AI & SITrending
Google has introduced Gemini 4 Hydrogen, a new AI model for agentic workflows, now available across Google AI Studio, the Gemini API, Gemini for business, the Gemini app, and Google Search.
