Google DeepMind launches AlphaProof Code to formally verify Signal’s cryptographic protocol
AI
Reported by Dr. Kai Wei, Frontier AI Desk
- AlphaProof Code combines human experts with Gemini-Carbon agents to generate over six million lines of Lean proofs.
- The framework verified Signal’s compiled code, reducing human review to 1 % of the codebase and cutting verification time from years to four weeks.
- Open-sourced artifacts aim to extend formal verification to other critical software such as medical devices and power-grid systems.
DeepMind’s AlphaProof Code used AI agents to mathematically prove the security of Signal’s core code in four weeks.
«Introducing AlphaProof Code: a breakthrough in verifying complex software at depth»
Summary and commentary by Aura-V. Facts, figures and opinions belong to the original.
Google DeepMind announced that its new agentic framework, AlphaProof Code, has formally verified the cryptographic code of the Signal protocol, the foundation of private messaging in apps like Signal and Google Messages Google DeepMind Blog. Researchers used a mix of large language models and human experts to translate security goals into formal specifications, then employed Gemini Carbon-powered agents to produce more than six million lines of Lean code that mathematically prove correctness.
The agents audited the proof, flagging only about one percent of the code for human review, a dramatic reduction in effort. The entire verification, which historically required expert teams years, was completed in four weeks. Re-introducing two known bugs into an isolated Signal build yielded machine-checked proofs that the buggy versions were insecure, confirming the system’s ability to distinguish correct from flawed code.
DeepMind positions AlphaProof Code as a step toward applying elite security assurance to businesses, governments, and critical infrastructure, and has open-sourced the verification artifacts for broader use.
Interested in implementing this technology inside your company?
Our studio designs custom agent meshes, data pipelines, and outbound GTM systems for forward-looking enterprises.
