DeepMind Introduces AlphaProof Code

Google DeepMind has introduced AlphaProof Code, an extension of its AlphaProof system now targeting real-world software verification. The announcement, dated November 10, 2026, describes the tool as capable of formally verifying complex software systems — with Signal's end-to-end encrypted messaging protocol cited as a notable use case. Formal verification uses mathematical proofs to confirm that software behaves exactly as specified, catching bugs and security flaws that conventional testing misses. AlphaProof Code builds on DeepMind's earlier work in AI-assisted mathematical reasoning, applying similar techniques to code correctness. The system leverages Gemini models to translate software properties into formal proof obligations and then attempts to discharge those proofs automatically. Verifying a widely deployed security protocol like Signal's represents a significant step up in complexity from prior benchmarks. The blog post frames this as a 'breakthrough,' though the full technical paper and scope of verified properties have not yet been detailed in the announcement text available.