Google DeepMind announced on November 10, 2026, that its AlphaProof Code framework had mathematically proved selected security properties of Signal’s compiled cryptographic code. According to The Register, the project took four weeks for a team of five people and generated 6.5 million lines of Lean proof code.
The distinction is between proving specified properties and declaring an application secure. For small software vendors facing buyer security reviews, the work offers a reference point for precise evidence, not a basis for claiming their own products are formally verified.
What the proofs cover
In its announcement, “Introducing AlphaProof Code”, DeepMind says human experts worked with models to translate security goals into mathematical specifications and reviewed every specification. Agents powered by Gemini Carbon generated proofs, which Lean checked. An agent auditing pass flagged one percent of the codebase for further human review.
“We don’t want this sort of vibe world where AI is generating the goals and AI is generating the proofs and basically taking tests that are self-written,” Swarat Chaudhuri, DeepMind research scientist, told WIRED.
The accompanying research paper, “AlphaProof Code: Verifying Signal’s Security at the Machine-Code Level”, describes proofs concerning Signal’s post-quantum PQXDH handshake and Triple Ratchet. These cover forward secrecy, authentication, post-compromise security and handshake secrecy. The proofs relied on approximately 26,000 lines of trusted specifications and AArch64 semantics, the paper says.
WIRED reports that the analysis found no new vulnerabilities in the portions examined. DeepMind also tested the framework by reintroducing two previously fixed bugs into an isolated version of Signal’s code. In both cases, DeepMind says, the system produced machine-checked proofs that the altered version was insecure.
Those results have explicit boundaries. The project didn’t formally verify Signal’s entire mobile app, WIRED reports. The Register adds that the result assumes perfect cryptography and doesn’t guarantee computational secrecy of bytes.
Research, not a purchasable assurance
DeepMind says it is open-sourcing the AlphaProof Code formal verification artifacts for community use. But Axios describes the work as research rather than a commercially available cybersecurity product. According to WIRED, DeepMind performed the analysis free of charge as a research project; Google didn’t disclose its cost or comment on plans for a commercial verification offering. DeepMind itself says scaling the techniques requires continued research.
For a small-team software company’s owner or founder, that leaves no established basis for treating comparable verification as affordable or commercially available.
The practical implication concerns the security evidence software buyers may expect as AI cyber capabilities advance. Security assurance can influence whether a prospect advances a deal. A supportable response is to document the controls, tests and independent assessments a product actually has, alongside their scope and limitations.
The Signal research illustrates why that scope belongs in the claim itself: which code was examined, which properties were proved, and which assumptions the proof still depends on.