# DeepMind uses AI to formally verify parts of Signal protocol

> Google DeepMind's AlphaProof Code proved key security guarantees of Signal in a four‑week proof‑of‑concept project.

Oossa · 2026-11-10 · https://oossa.com/en/deepmind-uses-ai-to-formally-verify-parts-of-signal-protocol

Google DeepMind researchers have shown that an AI system can mathematically prove important security properties of the Signal messaging protocol. The team used a new framework called AlphaProof Code to formally verify several cryptographic guarantees. The work took about four weeks and was done as a research project, not a commercial product. No new bugs were found in the sections they examined, and the AI correctly flagged older, already‑fixed flaws when it was run on previous versions of the protocol.

Formal verification means using math to check that software behaves exactly as intended, leaving no room for hidden vulnerabilities. Historically this process required experts to write detailed specifications, translate them into formulas, and then construct the proof—a labor‑intensive task that could take months. By letting an agentic AI (an AI that can set and pursue its own sub‑goals) handle much of the heavy lifting, DeepMind cut the time down to weeks. The researchers say the AI still needed human oversight at key steps, but it could suggest and self‑correct proofs on its own.

Signal’s chief technology officer, Ehren Kret, noted that prior verification work covered only small code fragments. DeepMind’s approach, he said, allows “significant expansion” of what can be checked. The team is releasing the specification and proof they built so other groups can build on it. While the result is promising, DeepMind cautions that scaling this to entire apps or whole software supply chains remains a challenge, and the cost of such AI‑driven verification has not been disclosed.

## Why this matters

If AI can reliably prove the security of cryptographic protocols, everyday users of Signal, Google Messages, WhatsApp and similar apps could gain stronger confidence that their messages stay private. For businesses that depend on encrypted communication—banks, healthcare providers, and journalists—the technology could eventually provide legally verifiable guarantees that their software has no critical flaws.

## The facts

- The DeepMind project lasted about four weeks, according to the researchers.
- AlphaProof Code correctly identified known, already‑fixed vulnerabilities in older Signal versions.
- Signal’s CTO Ehren Kret said the new AI approach enables “significant expansion” beyond earlier verification efforts.
- The researchers released the specification and proof they created for public use.
- Google declined to share the cost of the analysis and has not announced a commercial product.

## Why it matters

For the average person, the news means that the encryption protecting their chats is now backed by an additional mathematical check, reducing the chance of hidden bugs. For companies that must prove compliance with privacy regulations, AI‑assisted verification could become a new way to demonstrate security without lengthy manual audits.

## Sources & references

1. [A New AI Tool From Google DeepMind Proved the Signal Protocol Is Secure](https://www.wired.com/story/a-new-ai-tool-from-google-deepmind-formally-verified-the-signal-protocol/) – Wired, 2026-11-10

Last updated: 2026-11-10