Will the Hodge Conjecture be resolved before 2027?

- [2026-10-14] I found no source showing the Hodge Conjecture in its full CMI form as proved, disproved, or officially recognised as resolved. None of the three criteria (peer-reviewed publication, authoritative-body statement, or two Source Agencies reporting a resolution) is met in anything I found. https://www.claymath.org/millennium/hodge-conjecture/ - [2026-10-07] OfficeChai reported that OpenAI released a preprint, "The rational Hodge conjecture for CM abelian varieties", dated 2026-09-30. It is a partial result for CM abelian varieties in all dimensions, not the general conjecture. The report says the paper is not peer reviewed, independent verification is pending, and the paper names OpenAI rather than individual authors as its author. https://officechai.com/ai/hodge-conjecture-abelian-varieties/ - [2026-10-06] A commenter on Tao's blog linked the OpenAI preprint on GitHub. The commenter said the earlier Twitter hype about a counterexample "was wrong". This is an anonymous comment, not a confirmed fact. https://terrytao.wordpress.com/2026/09/12/the-status-of-the-hodge-conjecture/ - [2026-09-12] Claire Voisin published a guest post on Tao's blog titled "The status of the Hodge conjecture". It describes the general conjecture as open. It cites Markman's recent proof for abelian fourfolds as big progress, and notes that Voisin's own counterexample covers the Kähler extension only. https://terrytao.wordpress.com/2026/09/12/the-status-of-the-hodge-conjecture/ - [2026-09-11] Tao's companion post "On the Hodge conjecture" says progress has been limited. It says the first open case, codimension-2 cycles on 4-folds, still seems far out of reach. It also says Markman's 2025 proof for abelian varieties of dimension 4 and 5 is an exciting recent development. https://terrytao.wordpress.com/2026/09/11/on-the-hodge-conjecture/ - [~2026-09-09] A viral X post claimed the Hodge conjecture was "very close to being verified by OpenAI". The Vocal article reports that CMI and OpenAI issued no statement and that no paper, code, or confirmation existed at the time. Explainx (about 2026-10-04) lists Hodge as an unverified rumour and notes OpenAI's NYT statement did not name it. https://vocal.media/fyi/the-hodge-conjecture-rumor-openai-a-viral-tweet-and-the-math-worlds-quiet-fury - [2026-09-08] Quanta reported that OpenAI mathematicians proved a finite-time blowup for Navier–Stokes (a forced variant). This is the separate Millennium problem that is contested. Wikipedia says OpenAI does not intend to claim the prize. That result is the context for the Hodge speculation. https://www.quantamagazine.org/ai-has-solved-one-of-maths-1-million-millennium-prize-problems-20260908/ - [2026-10-12] Tech-insider reports that CMI said Navier–Stokes has "apparently been settled" but has not declared it solved. It also says CMI rules require publication in a qualifying venue plus about two years of scrutiny before a prize. That is a secondary report. If it applies to Hodge, it makes official CMI recognition before 2027 very unlikely. https://tech-insider.org/ai-millennium-problem-manifold-odds-eoy-2027 - [2026-10-14] The CMI Hodge page and its list of Millennium problems still show only Poincaré as solved. The lecture-series page also says only Poincaré has been resolved. Wikipedia likewise lists only Poincaré as officially declared solved. https://en.wikipedia.org/wiki/Millennium_Prize_Problems - [2026-01-04, 2025-07] Crank and low-quality "proofs" circulate, for example Medium posts and an arXiv paper on K3 surfaces. A K3 result would only cover a special case, and none of these has been recognised. https://arxiv.org/abs/2507.18999 - [2026-08] An August 2026 arXiv paper reportedly says the Hodge conjecture remains open beyond the computational range it covers. This comes via the Vocal article, and I did not verify it. https://vocal.media/fyi/the-hodge-conjecture-rumor-openai-a-viral-tweet-and-the-math-worlds-quiet-fury BOTTOM LINE: As of 2026-10-14 the Hodge Conjecture is unresolved. The only fresh development is OpenAI's partial, unrefereed result on CM abelian varieties, and experts such as Voisin and Tao still treat the general case as open. A full resolution would need a new general claim to be published or accepted by the field within about 11 weeks, and I found none. NOT FOUND: Any CMI, AMS or IMU statement on the Hodge Conjecture. A full proof or counterexample claim from a credible source. A peer-reviewed publication of the OpenAI preprint. Any scheduled event before 2027-01-07 that would resolve the question. I also did not open the OpenAI GitHub paper itself.

- [2026-10-06] OpenAI withdrew its preprint "The rational Hodge conjecture for products of K3 surfaces" (dated Oct 4). The notice cites a sign error in an earlier "stabilization-trace" construction, first used in a paper on Weil classes on split abelian eightfolds. It says the withdrawal concerns the proof and does not claim the statement is false. This is a confirmed retraction in the field's current AI-driven push. https://github.com/openai/math/blob/main/preprints/The-rational-Hodge-conjecture-for-products-of-K3-surfaces-October-4-2026/README.md - [2026-10-07] OfficeChai reports OpenAI posted a paper, dated Sept 30, 2026, claiming the rational Hodge conjecture for CM abelian varieties. It describes this as a partial result for a special class, not the general conjecture. Its author is listed as "OpenAI". The paper is a preprint and is not peer-reviewed. https://officechai.com/ai/hodge-conjecture-abelian-varieties/ - [2026-09-22] FourWeekMBA summarises the chain of claims. OpenAI said it made "substantial progress" on an unnamed Millennium problem. An anonymous X post (~Sept 9) and The Information (Sept 17, anonymous sources) attributed this to Hodge, with staff expecting a solve "relatively soon". As of Sept 17 the Clay Institute had no paper or statement. These are rumours and anonymous reports, not confirmed. I could not fetch the page (403), so this comes from the search snippet. https://fourweekmba.com/ai-openai-hodge-conjecture-clay-scoreboard/ - [2026-09-12] Claire Voisin's guest post on Terence Tao's blog describes the general conjecture as having no good structural handle. The smallest open case is codimension 2 on fourfolds. She cites Markman's 2025 proof for abelian fourfolds as big recent progress, which still covers only a special class. Tao's Sept 11 post says the same: general dimension-4 varieties look "far out of reach". Experts therefore see no general proof. https://terrytao.wordpress.com/2026/09/12/the-status-of-the-hodge-conjecture/ - [2026-10-06] A reader comment on Tao's blog says the OpenAI CM-abelian-variety paper has appeared. It adds that Twitter hype about a counterexample to Hodge "was wrong". This is an unverified anonymous comment. https://terrytao.wordpress.com/2026/09/12/the-status-of-the-hodge-conjecture/ - [2026-09-27] Reference class: only 1 of the 7 Millennium problems, Poincaré, has been officially resolved, over 26 years since 2000. Perelman posted in 2002–03 and the Clay prize was awarded in 2010, about 7–8 years later. The Clay Institute itself says only Poincaré is resolved. https://www.claymath.org/events/millennium-prize-problems-lecture-series/ and https://en.wikipedia.org/wiki/Millennium_Prize_Problems - [2026-09-27] Clay requires peer-reviewed publication and only considers a proposed solution at least two years after publication. Martin Bridson, Clay's president, is cited for the peer-review requirement. That makes a Clay prize before 2027 effectively impossible. The contract's other routes remain open: a peer-reviewed article in a Source Agency journal, an authoritative body's written recognition, or two Source Agencies reporting it as solved. https://en.wikipedia.org/wiki/Millennium_Prize_Problems - [2026-09] The contract's third route (two media reports of a solve, with no reported retraction or unaddressed error) is the most plausible way for a YES. A Millennium problem, Navier–Stokes, already has a September 2026 OpenAI-presented proposed counterexample that is unverified and disputed over priority. It shows AI-lab claims can reach the press before verification. Wikipedia reports OpenAI said it does not intend to claim the prize. https://en.wikipedia.org/wiki/Millennium_Prize_Problems - [2026-10-14] The Clay page still lists Hodge as unsolved. It says the conjecture is known in special cases, such as solution sets of dimension under four, and is unknown in dimension four. https://www.claymath.org/millennium-problems/ - [2026-09-18] A Vocal.media piece reports an August 2026 arXiv paper saying Hodge remains open beyond a computational range, with "three independent obstructions". It also notes Hodge was "open" as of Feb 2026 and that earlier 2026 rumours came to nothing. It is a low-quality source. Crank "proofs" exist on arXiv and ResearchGate, and none has been accepted. https://vocal.media/fyi/the-hodge-conjecture-rumor-openai-a-viral-tweet-and-the-math-worlds-quiet-fury BOTTOM LINE: Historically about 0 of 6 remaining Millennium problems have been resolved in any given year, and the one resolution took years of vetting. Experts call the general Hodge conjecture far from reach. OpenAI's partial results are the main tail risk: its Sept–Oct 2026 claims have been partial, and one was already withdrawn. A full resolution accepted by the contract's criteria before 2027 would be a very large departure from the base rate. NOT FOUND: Any published expert-probability estimate or poll for a Hodge resolution by 2027. Any other prediction market's price on Hodge (I was told not to look for this contract's own price). A Clay Institute statement on the OpenAI claims. A peer-reviewed publication of any claimed full proof or disproof. A full read of the FourWeekMBA and The Information pieces.

YES means: the Hodge Conjecture, in its official Clay formulation, is proved, disproved or officially recognised as resolved before 2027 (via a peer-reviewed article, a written statement from an authoritative body, or two Source Agencies reporting a resolution), and no retraction occurs before the Expiration time. - [2026-10-06/09] OpenAI posted 722 AI-written manuscripts (372 families) on GitHub from a model posed about 4,000 problems, at roughly three hours of compute per result. The README singles out the CM abelian varieties Hodge proof as an exception to the standard procedure. It says some unformalized results "could have issues". The scale and pace of this output is the main YES-side argument: AI-lab output is compressing timelines far beyond past norms. https://github.com/openai/math - [2026-10-08] Altman's summary of the release named four headline results. The Hodge item, for CM abelian varieties, has no Lean proof yet. The other three do. Lean formalization is the likeliest route to quick credibility, so a Hodge Lean certificate would be an early upgrade signal. https://officechai.com/ai/openai-releases-ai-generated-proofs-of-quasi-riemann-hypothesis-unique-games-conjecture-hodge-conjecture-for-cm-abelian-varieties/ - [2026-10-08] OpenAI says it began large-scale work on all Millennium problems on Sept 1. It has since posted a Navier–Stokes claim with Lean code (Sept 8). Hodge-specific output followed within weeks (Sept 30 CM paper, Oct 4 K3 paper). The K3 paper was withdrawn on Oct 6 for a sign error, which shows OpenAI is attacking Hodge-related cases in increasing generality. Next would be larger classes (abelian varieties, fourfolds) rather than a single jump to the general case. https://en.wikipedia.org/wiki/Navier%E2%80%93Stokes_priority_controversy - [2026-10-12] A Tao blog commenter argued that an AI company was more likely to disprove Hodge than prove it. The suggested route was Kontsevich's tropical-degeneration idea, where counterexamples might be found by large computations. This is an unverified comment. It is nonetheless the most concrete mechanism by which a disproof could appear quickly, since a single counterexample plus a machine-checkable certificate is far easier to verify than a proof. https://terrytao.wordpress.com/2026/09/12/the-status-of-the-hodge-conjecture/ - [2026-08] An arXiv survey ("GPT, the Counterexample Machine") lists ten conjectures AI reportedly disproved in 2026. These include the Jacobian conjecture in dimension 3 and the Erdős unit-distance conjecture, so counterexamples to long-standing conjectures have recently become achievable. It says nothing specific about Hodge. https://arxiv.org/pdf/2608.29595 - [2026-09-17] Per The Information (via Superpower Daily), OpenAI staff expected a Hodge claim "soon". The same piece says OpenAI might delay it to improve presentation after the Navier–Stokes backlash. This is a single anonymous source, but it implies a general claim is in preparation and may be timed for release. https://superpowerdaily.com/posts/openai-may-delay-a-reported-hodge-conjecture-result-after-math-backlash - [2026-09-22] OpenAI told the NYT it had made "substantial progress" on another unnamed Millennium problem. It was reported as announced and not yet disclosed. Decrypt noted that independent confirmation has historically taken years. This is the main bar to the official routes. https://decrypt.co/378551/openai-progress-second-millennium-prize-math-problem - [2026-10] The most plausible path to YES is the third route (two Source Agencies reporting the conjecture as resolved by a specific public work, with no reported unaddressed error). The prerequisites are: 1. OpenAI or a rival posts a full proof or disproof, probably with a Lean certificate. 2. Two or more outlets (Nature, Quanta, NYT, Wired and the like) describe it as resolved before Jan 7, 2027. 3. No outlet reports a retraction or unaddressed error. This is the hardest prerequisite, given the Oct 7 withdrawal of three manuscripts and the Oct 6 arXiv paper finding discrepancies between the Lean and natural-language Navier–Stokes proofs. Sources: https://aiweekly.co/alerts/openai-posts-722-ai-generated-math-preprints-pulls-three and https://en.wikipedia.org/wiki/Navier%E2%80%93Stokes_priority_controversy - [2026-10-09] The Oct 6 batch contains no complete solution to any of the five remaining Millennium problems, only partial progress on Riemann and Hodge. The CM abelian result is rational Hodge for one class, not the Clay statement. A literal reading of the Clay formulation would not count it. https://aiweekly.co/alerts/openai-posts-722-ai-generated-math-preprints-pulls-three - [2026-09-11] CMI president Bridson called the AI announcements "exciting" but stressed the peer-review requirement. CMI's own Navier–Stokes announcement is dated Sept 11. The contract's second route (a written recognition by an authoritative body) requires only a written statement by CMI, AMS or IMU referencing a specific work. The AMS published a leadership statement on Navier–Stokes on Sept 8, which shows these bodies are responding within days. https://en.wikipedia.org/wiki/Navier%E2%80%93Stokes_priority_controversy BOTTOM LINE: The strongest YES case is a surprise full AI proof or counterexample, with a certificate, covered by two outlets before Jan 7 and not withdrawn. It is supported by OpenAI's Hodge-specific work since Sept 30 and by press reports that a claim is being prepared. Against that, the Oct 6 releases were partial and the K3 paper was already withdrawn. Experts regard the general case as out of reach, and the other routes need peer review or scrutiny that cannot finish in about 11 weeks. NOT FOUND: Any full Hodge proof or counterexample claim. A Lean certificate for the Hodge result. A CMI, AMS or IMU statement on Hodge. Any peer-reviewed publication of an OpenAI Hodge paper. A confirmed date for OpenAI's rumored Hodge announcement. The text of the CMI Sept 11 Navier–Stokes announcement, which I did not open.

YES means: the CMI-formulation Hodge Conjecture (the general conjecture) is proved or disproved, and one of the three contract routes is met before the deadline. Those routes are a peer-reviewed journal article, a written recognition by an authoritative body, or two Source Agencies reporting a resolution with no retraction or unaddressed error. Strongest NO case: - [2026-10-06/08] OpenAI's Oct 6 release was 722 manuscripts, and the catalogue lists its Hodge result as "CM abelian varieties" only. The Lean proof for that result was "not yet provided" (BigGo, Oct 7). OpenAI's catalogue states the result as proved, but Altman described the headline claims as not yet confirmed by outside mathematicians (OfficeChai). So even OpenAI's own best Hodge claim covers a special class and is unchecked. https://officechai.com/ai/openai-releases-ai-generated-proofs-of-quasi-riemann-hypothesis-unique-games-conjecture-hodge-conjecture-for-cm-abelian-varieties/ - [2026-10-11] The retraction rate is a mechanism against YES. OpenAI withdrew 3 manuscripts on Oct 7 (a sign error) and revised 14 others (Retraction Watch, via CellCog). An OpenAI spokesperson said about 50% of results went out unconfirmed, and the README warns that unformalized results "could have issues". The K3 Hodge preprint was among the withdrawn. The withdrawal clause in the third route therefore bites for any AI-lab Hodge claim. https://cellcog.ai/blog/openai-math-results/ - [2026-10-14] The CMI rules, which I fetched, require publication in a Qualifying Outlet, then at least two years, then general acceptance. They also say CMI accepts no direct submissions. CMI cannot recognise a Hodge solution before 2027. The first full year-end cycle also cannot complete, because journal refereeing alone usually takes longer than 11 weeks. Route 1 and the CMI part of Route 2 are therefore effectively closed. https://www.claymath.org/millennium-problems/rules/ - [2026-10-14] CMI's own wording on Navier–Stokes, as reported by MathWorld, shows how it treats AI claims even when it is positive. On Sep 11 it called the problem "apparently" settled and awarded nothing. It still labels the problem "Active" and says credit follows its prize rules. Hodge, which has no claimed general proof, would get no softer treatment. https://mathworld.wolfram.com/MillenniumPrizeProblems.html - [2026-09/10] Recent genuine Hodge progress is all in special cases. These include Markman's abelian varieties of dimension ≤5 (van Geemen–Rapagnetta, arXiv 2607.18341), Floccari–Fu (J. Math. Pures Appl., June 2026), and a Fermat fourfold census (arXiv 2608.18134). That fits the expert view that the general case needs a new idea, not a finite extension. https://arxiv.org/pdf/2607.18341 - [2026-03] An arXiv paper (2603.20268) on abelian sixfolds notes that Markman's method does not extend even to the next case, McMullen's curve. That is evidence that incremental extension of current methods stalls before dimension 6. https://arxiv.org/pdf/2603.20268 - [2026-09/10] The reported "imminent Hodge solve" rests on one anonymous OpenAI source cited by The Information, plus a viral X leak. Superpowerdaily notes the company was reportedly weighing how to present it after the Navier–Stokes backlash. No general Hodge claim has appeared in the 3+ weeks since those reports; only the partial CM result and its withdrawn K3 sibling did. https://superpowerdaily.com/posts/openai-may-delay-a-reported-hodge-conjecture-result-after-math-backlash - [2026-10-08] The community backlash lowers the chance that authoritative bodies endorse a new claim quickly. The Association for Human Mathematics urged mathematicians to stop working with OpenAI. The IAS-hosted Advisory Group's guidelines (Sep 29) push for community-hosted release, which OpenAI has not met. A disputed priority fight (Buckmaster) already shadows its Navier–Stokes claim. https://cellcog.ai/blog/openai-math-results/ - [2026-10-14] Route 3 technicality: the claim must resolve the conjecture "in its original form". A CM-abelian or K3-product result is a subcase, so media calling it "partial" does not count. Even a counterexample would need to be to the integral or rational Hodge conjecture as the CMI states it. Voisin's Kähler counterexample and the integral-coefficient counterexamples are already known and do not count. Wire-style reports that overstate a partial result are a residual risk, but this is limited by the requirement that two Source Agencies describe the conjecture itself as resolved. https://www.claymath.org/millennium/hodge-conjecture/ - [2026-10-14] I found no event scheduled before 2027-01-07 that could trigger a resolution. Those would be an ICM, AMS or IMU announcement, or an Annals or Inventiones acceptance of a general proof. Status-quo inertia favours NO by default. BOTTOM LINE: Every route requires a general proof or counterexample that does not yet exist in credible form. The CMI two-year rule and the Oct 7 withdrawals make official or error-free recognition within 11 weeks very unlikely. The only plausible YES path is a sudden, unverified AI-lab claim of the general conjecture that two outlets report and that survives unretracted to Jan 7. NOT FOUND: Any claim of a general proof or counterexample from a credible author. Any AMS, IMU or national academy statement on Hodge. Retraction Watch's article itself (I saw it only via CellCog). The OpenAI GitHub CM-abelian paper itself and any expert referee verdict on it. Any scheduled journal acceptance.