DeepMind uses AlphaProof Code to formally verify Signal messaging security
Original title: Google DeepMind melds humans and AI agents to vet Signal's secure messaging techniques
Human-guided agents produced a Lean proof of libsignal’s PQXDH and Triple Ratchet, checked by the Lean kernel.
Google DeepMind researchers used AlphaProof Code, a multi-agent formal specification and verification framework, to check Signal’s post-quantum PQXDH handshake and Triple Ratchet. A human-guided agent wrote a Lean specification; Gemini Carbon agents generated Lean proofs of libsignal’s forward secrecy, authentication, post-compromise security, and handshake secrecy. The team says the result assumes perfect cryptography and does not guarantee computational secrecy of bytes. The work took four weeks for five people after agents generated 6.5 million lines of Lean code.
Unconfirmed: DeepMind claims this is the first AI-assisted proof linking Signal’s protocol guarantees to a widely used implementation.
- Agents generated 6.5 million lines of Lean code, then flagged parts for human review.
- The proof covers forward secrecy, authentication, post-compromise security, and handshake secrecy.
- Researchers say the result assumes perfect cryptography and is not a secrecy guarantee for bytes.
- The Signal effort took four weeks with a team of five.
- It responds to an April Beneficial AI Foundation challenge to verify Signal in Lean.
- Vendors
- Products
- SignallibsignalAlphaProof Code
- Organizations
- Google DeepMindBeneficial AI Foundation
- AI models
- Gemini Carbon
From the original article
3 min read at theregister.comPost-quantum handshake and Triple Ratchet get a formal security check
By Thomas Claburn · theregister.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.