The development

What happened

Google DeepMind says AlphaProof Code, a multi-model AI-agent framework, proved specified security properties of code implementing Signal's cryptographic protocol. Humans reviewed the mathematical specifications, while AI agents generated proofs checked by Lean. This is a claim about selected code properties, not proof that the entire messaging application is unhackable.

Key facts

  • Researchers worked with models to translate security goals into formal specifications and reviewed every specification. Gemini Carbon-powered agents then constructed proofs, with Lean checking their validity.

  • DeepMind reports completing the verification in four weeks. That is the researchers' account of this project, not an independently timed comparison demonstrating the same speedup on arbitrary software.

  • In isolated copies of the code, researchers reintroduced two previously fixed bugs. DeepMind says the framework produced machine-checked proofs of insecurity in both cases.

  • The authors announce open-sourcing verification artifacts. TideReach has not independently audited the paper, artifacts, specifications or proof infrastructure; the report's scope is the public research announcement.

The context

Why it matters

AI agents could help reduce the manual effort of connecting cryptographic designs to implementation-level proofs. TideReach assessment: the specification remains central—machine checking shows a proof meets its formal statement, not that every relevant real-world threat was covered. Other application code, devices and operational risks are not cleared by this announcement.