AI
Google DeepMind's AI Formally Verifies Signal's Core Encryption, Marking New Frontier for Cryptographic Proofs
Researchers used agentic AI to mathematically prove that key parts of Signal's cryptography work as intended, demonstrating both a validation of widely-trusted encryption and a breakthrough in AI-assisted software verification.
Ars Technica
Google DeepMind researchers have used AI to mathematically prove that critical components of the Signal protocol—the encryption that underpins messaging on Signal, Google Messages, and WhatsApp—function exactly as designed. The achievement, completed in just four weeks as a proof-of-concept, marks a significant milestone in both cryptographic validation and AI-powered software verification.
What Was Verified#
The team employed a novel AI framework called AlphaProof Code to formally verify several of the Signal protocol's key cryptographic guarantees. Rather than relying on traditional security audits or testing, formal verification uses mathematics to establish with certainty that software behaves as intended and lacks critical flaws. The analysis of Signal's protocol components revealed no new vulnerabilities in the sections examined, and when researchers tested the system on older Signal versions known to contain security bugs, the AI correctly identified them and flagged that those versions could not be formally verified.
How AI Changed the Game#
Formal verification has long been considered an ideal but impractical solution to software security. Historically, the process required researchers to manually translate complex software specifications into mathematical formulas and then construct formal proofs—work so time-consuming that it rarely scaled beyond small systems. Agentic AI dramatically accelerates this final stage, allowing analyses like DeepMind's to be completed in weeks rather than months or years.
However, the scope of this particular verification was limited; it did not, for instance, formally verify the entire Signal mobile application end-to-end, only key cryptographic components.
Broader Questions Remain#
While the breakthrough demonstrates real progress, the practical potential of AI-assisted formal verification remains uncertain. Security experts continue to debate whether the approach can scale effectively to the massive codebases that power modern software and whether it will become a standard practice in the industry.
Swarat Chaudhuri, a DeepMind research scientist, acknowledged the early-stage nature of the work: "We are very excited by this result, but there is a long distance to go before we are able to scale this to all the software of the world, and we want to do that."
DeepMind has made the Signal protocol specification and proof publicly available to support further formal verification research.

