Google DeepMind used AI to test Signal’s encryption: Here’s what it found

Researchers at Google DeepMind employed AI to perform a formal verification of Signal's encryption protocol, a process that proved the messaging app's security measures. Over a four-week period, DeepMind's AlphaProof Code framework scanned the encryption protocol for vulnerabilities but found none. This milestone in cryptography and AI validates the encryption used by platforms including Signal, WhatsApp, Facebook Messenger, and Google Messages.

DeepMind's application of AI to security assessments could expedite independent security checks, particularly in the face of rising AI-driven cyber threats. Signal's chief technology officer, Ehren Kret, expressed excitement about the result, while acknowledging the long journey still ahead to scale this method to all software. The encryption protocol, open-source and used by major messaging apps, ensures end-to-end encryption without the need for public key interception.

DeepMind researchers released their findings to aid other formal verification efforts, but emphasized that their scope was limited to older versions of the Signal protocol. They noted that human-AI collaboration would be necessary for future verification processes.