Google DeepMind's AlphaProof Code verifies Signal encryption in Lean

What DeepMind announced

Google DeepMind announced AlphaProof Code on 10 November 2026, describing it as a system for verifying complex software at depth. According to DeepMind, agents running on the system wrote three million lines of proofs in Lean, a formal proof language used to check that software behaves exactly as specified.

The stated target was Signal's core encryption, the cryptography that protects messages on the popular private messaging app. DeepMind published the announcement on its own blog, and no independent verification is yet reported.

Why formal verification matters

Formal verification means proving mathematically that code has no errors in the properties being checked, rather than relying on testing alone. Lean is a proof assistant, a tool that checks each step of such proofs automatically.

If DeepMind's claims hold, using AI agents to produce millions of lines of proofs could make deep verification of critical security code faster and cheaper. That would matter for messaging apps, banking software and other systems where encryption flaws carry serious consequences.

What to watch next

The announcement so far is limited to a short blog excerpt, so details about the methods, tooling and peer review are not yet available. Readers should treat the scale and depth of the verification as DeepMind's own claim until independent researchers or Signal confirm the results.