AivexaNewsSearch
AI news for builders and product teamsChecked every hour

Google Deepmind's AlphaProof Code verified Signal's cryptographic core in four weeks instead of years

Collected Nov 11, 2026

Google DeepMind has introduced AlphaProof Code, an AI system it says can formally prove that software meets its security specifications without errors. The company demonstrated it on Signal's encryption protocol, completing verification work in four weeks that DeepMind says would typically take years. The results were described in a blog post by DeepMind researchers Pushmeet Kohli and Swarat Chaudhuri.

What distinguishes this effort, per DeepMind, is that it targeted the actual cryptographic core code running on users' devices rather than stopping at the level of a theoretical design. Earlier verification efforts stopped at the theoretical design stage, according to the blog post. After human experts defined the formal security requirements, AI agents built on Gemini Carbon wrote more than six million lines in Lean, a proof language used for machine-checked mathematics. About one percent of the codebase was flagged for human review, and two bugs deliberately planted for testing were caught immediately.

In the workflow DeepMind describes, the AI agents help draft the specifications and then independently prove that the program code satisfies them. That division matters: formal verification normally requires a human to state precisely what "secure" means before any proof can be checked, and Lean's checker then accepts or rejects each proof mechanically. The agent contribution is in generating the proofs at scale, while the specification remains anchored in expert-defined requirements.

Signal CTO Ehren Kret said the approach strengthens trust in data privacy as AI threats grow, according to the article. DeepMind plans to extend the method to critical infrastructure such as power grids and medical devices through systems like CodeMender. This suggests the company sees formal verification of deployed code, not just reference designs, as a general-purpose offering; whether that extends cleanly to large, fast-changing codebases outside cryptography is an open question.

Why it matters: Formal verification has long been too slow and specialist-heavy for everyday software teams, so this result is a signal that proof generation may become a practical review step rather than a research project. For developers working on security-critical code, the near-term change is likely more machine-checked evidence alongside tests and audits, with human effort concentrated on writing specifications and reviewing flagged lines. Product teams in regulated or safety-critical areas should watch how far the method generalizes beyond a well-scoped protocol like Signal's.

Read at The Decoder

Based on reporting from the original publisher. Visit the source for full context and later updates.

Publisher excerpt

Google Deepmind's AlphaProof Code mathematically verified Signal's cryptographic core code in four weeks, a process that typically takes years. AI agents powered by Gemini Carbon wrote more than six million lines of proof code and flagged only about one percent for human review. Deepmind plans to extend the method to critical infrastructure like power grids and medical devices. The article Google Deepmind's AlphaProof Code verified Signal's cryptographic core in four weeks instead of years appeared first on The Dec