Google DeepMind's AlphaProof Code formally verifies Signal protocol cryptography

Google DeepMind announced on November 10, 2026, that its new AlphaProof Code system formally verified core cryptographic security properties of the Signal protocol's compiled implementation, the encryption layer that underlies both Signal and Google Messages, as described in the blog post introducing AlphaProof Code.

What changed

• AlphaProof Code is a multi-LLM agentic framework powered by Gemini Carbon, built for formal verification of software.
• The team used it to formally verify core cryptographic security properties of the Signal protocol's compiled implementation, the code underlying end-to-end encryption in Signal and Google Messages.
• The effort generated more than 6 million lines of Lean proof code in about four weeks.
• DeepMind says it is the first time AI has formally verified a widely deployed cryptographic protocol down to the compiled machine-code level.
• Automated review flagged only about 1% of the codebase as needing human review.
• In a validation test, the system correctly re-identified two previously fixed real bugs after they were reintroduced into isolated code versions.

What it means for AI builders

The result shows that AI agents can now formally verify production cryptographic code at a scale and depth that was previously impossible, a capability directly relevant to developers and security teams building security-critical software with AI assistance. It also points to new agentic verification tooling that could extend beyond Signal to other widely used codebases.

Availability

DeepMind open-sourced the formal verification artifacts and specification from the project for community use.

Sources

• Introducing AlphaProof Code: a breakthrough in verifying complex software at depth (Google DeepMind, November 10, 2026, official)

Written with AI assistance, strictly from the sources listed above. Before it went live, automated checks verified its dates, sources and every language edition.