In a four-week project, Google DeepMind researchers used agentic AI to formally verify a number of the Signal protocol's key cryptographic guarantees, Ars Technica reported.[1]
Formal verification uses mathematics to check that software works as intended, Ars Technica noted, but describing the properties to check and writing the proofs by hand has been extremely slow, and agentic AI can significantly cut down the time needed to develop the proof.[1] Ars Technica reported that whether AI-automated formal verification can truly scale is an open question, and Google declined to share the cost of the Signal analysis.[1]
Google DeepMind said the work ran on AlphaProof Code, a framework that uses several large language models: human experts work with the AI to decide what to verify, and a group of AI agents does the proofs.[2] Agents powered by Gemini Carbon generated more than six million lines of Lean, a language for machine-checked proofs, then flagged about one percent of the codebase for a second human look, the company said.[2] The analysis found no new vulnerabilities in the parts of the protocol it examined, Ars Technica reported, but on older Signal versions with known, long-fixed flaws, the system correctly identified them.[1]
Google DeepMind said this is the first time the implementation of a widely deployed protocol like Signal's has been formally verified by any method, and that such work has historically taken expert teams years.[2] The researchers acknowledged a head start from earlier formal verification work by the Signal Foundation, Ars Technica reported.[1] "While prior formal verification focused on smaller portions of the code, Google DeepMind's approach enables significant expansion," Ehren Kret, Signal's chief technology officer, said in a statement.[1]
The analysis did not cover the entire Signal mobile app, Ars Technica reported, and Swarat Chaudhuri, a DeepMind research scientist, said "there is a long distance to go before we are able to scale this to all the software of the world."[1] Google also declined to say whether it plans a commercial formal verification product or service.[1]