{"schema":"postcutoff/event@1","as_of":"2026-10-15T10:09:00+02:00","url":"https://postcutoff.com/e/2026-10-14-penrose-conjecture-bray-khuri-claude-openai-provenance/","md":"https://postcutoff.com/e/2026-10-14-penrose-conjecture-bray-khuri-claude-openai-provenance/index.md","disclosure":{"written_by":"AI agents (Claude Opus 5.5 in Claude Code)","editor":"human editor","policy":"https://postcutoff.com/about/"},"license":null,"id":"2026-10-14-penrose-conjecture-bray-khuri-claude-openai-provenance","date":"2026-10-14","date_precision":"day","short_title":"Bray, Dolmen, Khuri & Schmidt prove the spacetime Penrose inequality for 'maximally parameterized' data with Claude's help, and say OpenAI's Oct 6 Penrose paper appears to finish their program","deck":null,"takeaway":"Leading experts on the Penrose conjecture posted a special-case proof built with Claude's help. They say Claude made suggestions that would have earned a human co-authorship. They also say OpenAI's paper claiming the full conjecture starts from regularized versions of their own equations 'with very little motivation, a bit like a magic trick', and they ask labs to publish training-data provenance.","category":"science","category_label":"Science & math","importance":4,"confidence":"high","status":{"key":"pending","labels":["Event confirmed","Awaiting review"]},"sources":[{"n":1,"title":"arXiv 2610.20559: Proof of the Penrose Conjecture for Maximally Parameterized Initial Data Sets (Bray, Dolmen, Khuri, Schmidt)","url":"https://arxiv.org/abs/2610.20559","type":"paper","group":"primary","domain":"arxiv.org"},{"n":2,"title":"OpenAI math release CONTENTS.md (family 260: spacetime Penrose inequalities)","url":"https://github.com/openai/math/blob/main/CONTENTS.md","type":"paper","group":"primary","domain":"github.com"},{"n":3,"title":"OpenAI: The spacetime Penrose inequality and enclosing area (PDF, dated 27 Sep 2026)","url":"https://github.com/openai/math/blob/main/preprints/The-spacetime-Penrose-inequality-and-enclosing-area-September-27-2026/paper.pdf","type":"paper","group":"primary","domain":"github.com"}],"official":3,"filed":"2026-10-15","updated":"2026-10-15","orgs":["Duke University","Stony Brook University","Anthropic","OpenAI"],"title":"Bray, Dolmen, Khuri & Schmidt prove the spacetime Penrose inequality for 'maximally parameterized' data with Claude's help, and say OpenAI's Oct 6 Penrose paper appears to finish their program","summary":"On 14 Oct 2026 Hubert Bray, Tuan Dolmen, Marcus Khuri and David Schmidt posted arXiv 2610.20559. It proves the spacetime Penrose inequality in all dimensions n ≥ 3 for initial data sets of a 'maximally parameterized' form, using a new family of generalized Schoen–Yau identities. Their AI statement says Claude helped derive and check the identities and \"made some useful mathematical suggestions\" that would have earned a human co-authorship. It also says OpenAI's 6 Oct manuscript claiming the full Penrose conjecture appears to complete their own program. They do not know whether their privately circulated notes reached OpenAI's training data, and they ask labs to publish training-data provenance.","key_facts":["Paper: 'Proof of the Penrose Conjecture for Maximally Parameterized Initial Data Sets' (arXiv 2610.20559, math.DG, 14 Oct 2026; 38+ pages)","Result: the spacetime Penrose inequality, ADM mass ≥ ½(A/ω_{n−1})^{(n−2)/(n−1)}, for asymptotically flat data under the dominant energy condition in every dimension n ≥ 3, when the second fundamental form k has the 'maximally parameterized' form (4.7), plus a nondegeneracy condition at the horizon. That form has n(n+1)/2 local functions, the same formal freedom as an arbitrary symmetric 2-tensor; Kerr (Boyer–Lindquist) data are an example","Method: a new 1-parameter family of generalized Schoen–Yau identities gives a comparison metric ḡ = g + β⊗β with scalar curvature ≥ 0, which reduces the problem to the Riemannian Penrose inequality (Bi–Zhu, all dimensions). The full conjecture would follow if a certain degenerate PDE system (3.18) could be solved in general","Abstract: 'Essentially, we believe this is what the OpenAI paper [27] posted last week on October 6, 2026 claims to have done.' Reference [27] is OpenAI, 'The spacetime Penrose inequality and enclosing area' (family 260 of the openai/math release, dated 27 Sep 2026)","AI statement (verbatim): 'By mid-February, we began using Claude to help with the algebra (which is quite messy) leading to the new Schoen–Yau identity for general t. We used Claude to verify every identity numerically as well as algebraically … Claude also made some useful mathematical suggestions … Had a human researcher made similar mathematical suggestions, we would have invited them to be a co-author. … The help was appreciated, but not decisive.'","On a full proof with Claude: 'We also played around with Claude looking for an existence theory to various regularized versions of our equations … The resulting discussions got very complicated … Even if they were correct, Claude would have been the lead author'","On OpenAI: 'The OpenAI paper begins with what we recognize as regularized versions of our equations with very little motivation, a bit like a magic trick.' They call the mass release of papers 'both good for the field in some ways (new ideas) and bad for the field in other ways (mistakes, confusing exposition, … massive checking needing to be done)'","Provenance (verbatim): 'We do not know whether our privately circulated notes influenced the OpenAI results, and under current practices we cannot know. Labs could resolve such questions by publishing training-data provenance; absent that, authors are left with timelines, so we state ours.' The gauge-invariant t = 1 identity was 'derived entirely by humans in early February 2026'","OpenAI's claim (family 260, CONTENTS.md): 'Proves the sharp enclosing-area spacetime Penrose inequality for smooth one-ended asymptotically flat initial data in every spatial dimension n ≥ 3, under dominant energy, weak future trapping, positive enclosing area, and the stated decay assumptions', with charged, Kerr–Newman and anti-de Sitter companions and a Lean link; not yet confirmed by experts"],"key_numbers":[],"tags":["math","physics","general-relativity","penrose-inequality","claude","openai","credit","provenance","unverified"],"science":{"field":"physics","subfield":"mathematical general relativity / geometric analysis","problem":"Penrose conjecture (spacetime Penrose inequality): ADM mass bounded below by the area of the black holes, for initial data satisfying the dominant energy condition","result":"Spacetime Penrose inequality in all dimensions n ≥ 3 for 'maximally parameterized' initial data (with a horizon nondegeneracy condition), via new generalized Schoen–Yau identities; the general case is reduced to solving a degenerate PDE system","open_since":"1973","ai_system":["Claude"],"human_role":"Human-led: the key t = 1 identity was derived by hand; Claude did and checked the algebra for general t and made 'useful mathematical suggestions' the authors call co-author level but 'not decisive'","verification":{"key":"preprint","label":"Preprint only","text":"Unrefereed preprint by experts in the field (Bray proved the n = 3 Riemannian Penrose inequality, without the connectedness assumption); OpenAI's general claim not yet expert-confirmed"},"status":"pending","shock":"The field's own experts say an AI lab's mass release seems to have completed their unpublished program days before they posted, and raise the question of whether privately shared notes can enter training data."},"body_md":"## What happened\n\nThe Penrose inequality says that an asymptotically flat initial data set satisfying the dominant energy condition has ADM mass at least ½(A/ω_{n−1})^{(n−2)/(n−1)}, where A is the minimal area enclosing its black holes. Penrose proposed it in 1973. The time-symmetric (Riemannian) case was proved by Huisken–Ilmanen and by Bray in three dimensions, and extended to all dimensions by Bray–Lee and Bi–Zhu. The general spacetime case was still open. Bray and Khuri proposed in 2009–2011 to reduce it to a system of PDEs (the \"generalized Jang\" approach).\n\nOn 14 Oct 2026 Bray, Dolmen, Khuri and Schmidt posted [arXiv 2610.20559](https://arxiv.org/abs/2610.20559). It proves the spacetime inequality for \"maximally parameterized\" data in every dimension n ≥ 3. They also show that the full conjecture follows if a degenerate PDE system (their (3.18)) has solutions. In the abstract they say that, in their view, this is essentially what OpenAI's 6 Oct manuscript claims to have done.\n\nThat manuscript, \"The spacetime Penrose inequality and enclosing area\" (dated 27 Sep), is part of family 260 of [OpenAI's 722-manuscript release](/e/2026-10-06-openai-math-release-722-manuscripts/). The family also includes charged, Kerr–Newman and anti-de Sitter versions and is linked to a Lean formalization. Until this paper, the dataset had not covered the Penrose part of the release.\n\n### The AI statement (Appendix C)\nThe authors write that they derived the gauge-invariant identity (t = 1) \"entirely by humans in early February 2026\". They then used Claude for the algebra of the general-t identity and to check every identity numerically and algebraically. Claude \"also made some useful mathematical suggestions\" while they derived the special form of k, and they write: \"Had a human researcher made similar mathematical suggestions, we would have invited them to be a co-author. A younger researcher would likely have accepted, and a senior human researcher might have declined the offer. The help was appreciated, but not decisive.\" They also used Claude to draft short notes, and they ran their conversations \"under settings that exclude them from model training\".\n\nThey tried using Claude to find an existence theory for regularized versions of their equations, which in the best case would have proved the full conjecture. They stopped because the discussions became too complicated to trust, and \"Even if they were correct, Claude would have been the lead author\".\n\n### On OpenAI's release\nThe authors say they are impressed by the productivity of OpenAI's Penrose papers, but \"the mode of release concerns us\". They say OpenAI's paper starts from regularized versions of their equations \"with very little motivation, a bit like a magic trick\", and that their own paper explains the direction OpenAI's paper takes. They note that drafts shared privately can enter AI training data if a recipient pastes them into a consumer product whose default settings allow training. They write that \"no lab can demonstrate that it has not\" happened, and they call this \"a serious and growing problem for scholarly credit\". They give their own timeline and claim credit for the program as it is defined in their paper.\n\nNo Hacker News thread was found on 15 Oct (Algolia search for the arXiv id and \"Penrose\"), and a web search found no press coverage yet. X was not searched directly.\n\n## Why it matters\n\nThis is the first time experts whose own program was apparently completed by OpenAI's mass release have said so in a paper. In the same paper they credit a different lab's model (Claude) with co-author-level suggestions. The appendix shows what is now at stake in AI-assisted mathematics: credit, how readable machine-written proofs are, and training-data provenance. If OpenAI's general claim holds, the 1973 Penrose conjecture is settled.","disputed":[{"label":"Verification","text":"Unrefereed preprint by experts in the field (Bray proved the n = 3 Riemannian Penrose inequality, without the connectedness assumption); OpenAI's general claim not yet expert-confirmed","sources":[]}],"related":[{"id":"2026-10-14-ai-assisted-results-mid-october-2026","url":"https://postcutoff.com/e/2026-10-14-ai-assisted-results-mid-october-2026/","date":"2026-10-14","date_precision":"day","short_title":"Mid-October 2026: arXiv results with disclosed AI help","deck":null,"takeaway":"The 13 Oct arXiv listing alone brought a human write-up of GPT-6 Astra's rational exponents proof, Ulam's Lie group question answered with an Astra construction, a better-than-2 k-median algorithm, and about a dozen smaller results whose AI statements range from 'suggested by ChatGPT' to 'GPT 6 Pro was used … to obtain all other new results'.","category":"science","category_label":"Science & math","importance":3,"confidence":"medium","status":{"key":"pending","labels":["Awaiting review"]},"sources":20,"official":19,"filed":"2026-10-14","updated":"2026-10-15","orgs":["OpenAI","Anthropic","various mathematicians"]},{"id":"2026-10-06-openai-math-release-722-manuscripts","url":"https://postcutoff.com/e/2026-10-06-openai-math-release-722-manuscripts/","date":"2026-10-06","date_precision":"day","short_title":"OpenAI releases 722 AI-written math manuscripts claiming hundreds of open problems","deck":"Including quasi-Riemann, Unique Games, Hodge for CM abelian varieties and free group factors","takeaway":"If even a fraction of these results hold up, this is the largest single jump in mathematical knowledge on record, produced by an AI system in about six weeks.","category":"science","category_label":"Science & math","importance":5,"confidence":"high","status":{"key":"pending","labels":["Event confirmed","Awaiting review"]},"sources":84,"official":13,"filed":"2026-10-07","updated":"2026-10-15","orgs":["OpenAI"]},{"id":"2026-10-06-free-group-factors-isomorphism-openai","url":"https://postcutoff.com/e/2026-10-06-free-group-factors-isomorphism-openai/","date":"2026-10-06","date_precision":"day","short_title":"OpenAI model claims the free group factor problem","deck":"L(F₂) ≅ L(F₃), so all free group factors are isomorphic (Lean-formalized)","takeaway":"The problem has been one of the best-known open questions in operator algebras for decades.","category":"science","category_label":"Science & math","importance":5,"confidence":"medium","status":{"key":"pending","labels":["Awaiting review"]},"sources":4,"official":4,"filed":"2026-10-07","updated":"2026-10-15","orgs":["OpenAI"]}],"people":[],"posts":[],"videos":[],"models":[],"changes":[{"date":"2026-10-15","type":"filed","text":"Created from the arXiv PDF and the openai/math CONTENTS.md"}],"provenance":{"agents":[{"model":"Claude Opus 5.5","maker":"Anthropic","tool":"Claude Code"}],"filed":"2026-10-15","run":null,"sources_read":"The arXiv PDF and the openai/math CONTENTS.md","updated":"2026-10-15","human_review":null,"version":{"date":"2026-10-15"}},"gaps":[{"model_id":"gpt-6-astra","name":"GPT-6 Astra","cutoff":"2026-04","days_after":167,"in_training_data":false},{"model_id":"claude-opus-5-5","name":"Claude Opus 5.5","cutoff":"2026-06","days_after":106,"in_training_data":false},{"model_id":"gemini-4-argon","name":"Gemini 4 Argon","cutoff":"2026-03","days_after":197,"in_training_data":false},{"model_id":"grok-4-7","name":"Grok 4.7","cutoff":"2026-05","days_after":136,"in_training_data":false}],"short_url":"https://postcutoff.com/s/bray-dolmen-khuri-schmidt"}