What DeepMind announced
Google DeepMind's post, dated November 10, 2026, introduces AlphaProof Code, an agentic framework for formal verification. Researchers used it, DeepMind says, to mathematically prove key security properties of the actual code underlying the Signal protocol, which protects conversations in apps such as Signal and Google Messages.
The framework splits the work the way a research team would: humans work with AI to decide what to prove, and agents do the proving. DeepMind's blog says researchers first translated Signal's security goals into formal specifications, with humans reviewing every one, and that agents powered by Gemini Carbon then generated over six million lines of Lean. The technical paper linked from that post is more specific: it says the specifications were developed in a back-and-forth between humans and agents powered by Gemini 4 Carbon and Claude Opus 5.5, and that the proofs were produced entirely by agent teams powered by Gemini 4 Carbon.
Who can use it and when
AlphaProof Code is presented as a research result, not as a generally available product with a price or a self-serve signup. The announcement does not list public access, a waitlist, or commercial terms. DeepMind says it is open-sourcing formal verification artifacts and links a technical paper, Formal Verification of Signal, for readers who want the method in detail.
If you work on protocol verification, the practical next step is the paper and the published artifacts, not an assumption that you can point AlphaProof Code at your own codebase today.
Why it matters
Most protocol verification, DeepMind notes, stops at the design. The agents, it says, proved that compiled code meets specified properties, including keeping messages protected before and after a device is temporarily compromised, down to the code running on devices. That gap between design and implementation is the part of the claim that matters for security teams, provided the paper bears it out.
DeepMind says the agents audited the proofs and flagged about one percent of the codebase for a human second look, with Lean checking the rest, and that the verification took four weeks, a task it says historically takes expert teams years. Those time comparisons are the lab's, not an independent measurement. The researchers also say they reintroduced two previously fixed bugs into an isolated version of Signal's code; in both cases the system produced a machine-checked proof that the buggy version was insecure. Signal CTO Ehren Kret is quoted calling the work a significant expansion of prior, smaller-scope verification.
Access and limitations
There is no public product, region, or price in the announcement. The result is tied to a specific implementation and to properties the researchers specified. DeepMind frames the work as groundwork for other critical systems and mentions pairing it with defense agents such as CodeMender, which is a research direction, not a shipping guarantee.
The blog and the paper do not use the same model wording. The blog, dated November 10, 2026, says only Gemini Carbon. The paper, also dated November 10, 2026, says the proofs were produced entirely by agent teams powered by Gemini 4 Carbon, and that the specifications were developed with agents powered by Gemini 4 Carbon and Claude Opus 5.5. This article reports both formulations. A related-posts card on the blog that links a separate October 2026 article about Gemini 4 Carbon is not the evidence for that identification; the paper is.
Go to the source
Sources & further reading
Based on the official documentation below. Features and access can change; check the provider before making a decision.
Was this page useful?
Help us make our coverage more useful. This is feedback on the page, not a rating of the tool.
Go a little further