AI
Google DeepMind Uses AI to Mathematically Prove Signal's Encryption is Secure
Agentic AI helped researchers formally verify key cryptographic properties of the widely-deployed Signal protocol in just four weeks, validating the security of billions of encrypted messages.
WIRED — AI
Researchers at Google DeepMind have leveraged agentic AI to formally verify critical security claims about the Signal protocol, the cryptographic standard protecting communications across the Signal app, Google Messages, WhatsApp, and other platforms. The achievement demonstrates both the robustness of widely-trusted encryption technology and an emerging frontier in how artificial intelligence can strengthen software security.
What Was Accomplished#
The project, completed in roughly four weeks as a proof-of-concept, represents a significant milestone in combining human expertise with machine reasoning. Using a new AI framework called AlphaProof Code, the team mathematically proved that essential cryptographic properties of Signal's protocol work exactly as intended. No new vulnerabilities surfaced during the analysis of the portions examined, though when researchers tested the system against known flaws in older Signal versions, the AI correctly flagged them and demonstrated the protocol could not be formally verified with those bugs present.
The researchers have made their specification and formal proof publicly available to support future verification efforts.
Why This Matters for Security#
Formal verification represents a long-standing aspiration in software security: using mathematics rather than testing to definitively establish that code behaves correctly and lacks critical flaws. Traditionally this has been painfully slow—human specialists must articulate what they want to prove, translate those goals into mathematical formulas, and then construct the actual proof. Agentic AI dramatically accelerates this final stage.
Swarat Chaudhuri, a DeepMind research scientist, acknowledged the significance while tempering expectations: "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."
Current Limitations and Next Steps#
The current analysis was limited in scope—it did not formally verify the entire Signal mobile application, for instance. The researchers benefited from preliminary formal verification work the Signal Foundation had already undertaken, which gave them a head start.

