Google DeepMind AI Formally Verified Signal Protocol Security

Google DeepMind researchers used agentic AI to formally verify key cryptographic guarantees in the Signal protocol, a claim that matters because it points toward AI-assisted security that can reason about code with mathematical rigor. According to WIRED, the effort was a proof of concept using a new framework called AlphaProof Code and took about four weeks. The result does not mean every cryptographic claim can now be checked automatically. It is a single protocol, a limited set of guarantees, and a research project, according to the report. The development is being watched by people who build with AI, because it moves the conversation from generating outputs to checking formal properties. WIRED reported that the work is significant for the cryptography that protects communications on several major platforms.

What Happened at Google DeepMind

According to WIRED, Google DeepMind researchers used agentic AI to formally verify a number of Signal protocol cryptographic guarantees. The project was a proof of concept. The researchers worked with AlphaProof Code, an AI framework for formal verification. The reported timeline was four weeks. The work was significant in two broad ways, the article said: it offered an independent check on cryptography protecting large volumes of communications, and it showed a formal verification capability through a new AI approach. The researchers used agentic AI to mathematically prove, or formally verify, a number of key guarantees. The project was conducted as a proof of concept in four weeks. The WIRED report said the work is significant for the widely used cryptography and for AI itself.

The WIRED report noted that DeepMind conducted the work as a proof of concept in four weeks. The researchers said they were excited by the result but acknowledged a long distance to go before scaling to all software. Swarat Chaudhuri, a DeepMind research scientist, said in a statement reported by WIRED that the team was excited but wanted to scale the approach.

Why the Signal Protocol Matters

The Signal protocol is used across widely deployed messaging contexts, including Signal, Google Messages and WhatsApp, according to WIRED. Formal verification is meant to check that software meets intended behavior, which is especially important for encryption. DeepMind said the analysis did not find new vulnerabilities in the portions of the protocol it examined.

Formal verification has long been described as a holy grail in software security because it uses math to check and validate that software works as intended. The WIRED report said AI-aided formal verification is having a moment, but its practical use and effectiveness for large codebases remain uncertain.

What the AI Verified—and What It Did Not

The DeepMind researchers did not formally verify the entire Signal mobile app, the report said. The work focused on key cryptographic guarantees. The team also applied the method to older versions of Signal that contained known, long-fixed vulnerabilities. The researchers said AlphaProof Code correctly identified those cases and concluded that the protocol could not be formally verified in those versions. DeepMind said it planned to release the specification and proof from the Signal analysis.

The researchers emphasized that their findings are one milestone in a larger process. They said the Signal Foundation had done prior formal verification work, while DeepMind's approach enabled significant expansion.

What It Means for AI-Assisted Formal Verification

Formal verification has long been described as a difficult goal because it requires turning software requirements into mathematical proofs. Before AI, that process could be slow. Agentic AI was said to reduce the later stages of proof work. Still, the researchers said scaling to all software is a long-term goal. They also described a human-AI collaboration process that keeps human input and review at key points.

Pushmeet Kohli, chief scientist at Google Cloud and DeepMind's vice president of research, said that if correctness can be verified, companies could say they can prove properties. The open question is whether AI automated formal verification can truly scale and at what cost. Google declined to share the cost or comment on commercial plans. The researchers say some crucial advancements are in developing a human-AI collaboration process, as Chaudhuri puts it, that allows agents to contribute and self-correct at almost every phase while still seeking human input and review at key junctures. The WIRED report said the team did not want a vibe world where AI generates goals and proofs and basically takes tests that are self-written. There is a middle ground.

Where It Fits in a Grafy Workflow

For developers, the story touches formal verification, which is ultimately software proof work. Grafy Code is the AI coding studio in Grafy for agentic coding workflows, though the report does not mean Grafy Code now includes formal verification for cryptographic proofs. A developer who wants to explore code-level AI assistance could use Grafy Code. A researcher who wants to read source material and question it could use Grafy Reader. The story does not show that Grafy has a formal verification product. It does show a developer-oriented AI capability that could inform how coding assistants are evaluated.

Frequently asked questions

Did Google DeepMind use AI to formally verify the Signal protocol?

Yes. According to WIRED, Google DeepMind researchers used agentic AI and the AlphaProof Code framework to formally verify key cryptographic guarantees in the Signal protocol. The work focused on parts of the protocol, not the complete Signal mobile app. The company said the analysis did not find new vulnerabilities in the portions it examined.

Did the AI formally verify the entire Signal mobile app?

No. The DeepMind researchers told WIRED that their formal verification did not cover the entire Signal mobile app. They examined key protocol guarantees and also tested older versions containing known, long-fixed vulnerabilities. In those cases, AlphaProof Code correctly identified that the protocol could not be formally verified. The team described the result as one milestone, not a full verification.

Can AI automated formal verification scale to all software now?

Not yet. DeepMind researchers said there is a long way to go before scaling formal verification to all software. They described AI-aided formal verification as having a moment, while noting questions remain about practicality and cost. The company said the Signal project was free and research-focused, and it declined to discuss commercial plans.

---

Source: WIRED — AI — A New AI Tool From Google DeepMind Proved the Signal Protocol Is Secure