AkashicINQUIRY017 · 20 Nov 2026 · v1.0.0
Has AI Proved the Riemann Hypothesis?
A Proof Claim, Its First Checks and the Work Still Ahead
Claimed proof · Verification pending in the inspected record
The Question
Brian Greene has shared an extraordinary possibility: a proof of the Riemann Hypothesis developed with AI. What would establish that the problem has been solved, and what has actually been checked?
Current answer: there is an inspectable proof claim. In the verification record examined for this edition, the complete solution build and proof replay were still in progress. AkashicNET has not verified the Riemann Hypothesis.
From a Post to the Primary Claim
A note on mathematician Levent Alpöge’s website reports finding a proof while working with Claude. It also claims a generalised result and describes a further extension as an expectation. The release points readers to Lean source rather than a conventional explanatory paper. This is a substantive, checkable release, while the correctness of the argument remains a separate question. [S01] [S02]
The attribution matters. The primary announcement inspected is the note on Alpöge’s website. This review did not establish a current formal corporate announcement by Anthropic. It also cannot independently apportion the mathematical work between the human collaborator and the model.
What the Mathematics Asks
Prime numbers have an average distribution, with fluctuations around it. The Riemann Hypothesis concerns those fluctuations through the zeros of the analytically continued zeta function. It says the non-trivial zeros all have real part ½. Checking many zeros numerically cannot settle a statement about every non-trivial zero. [S12]
The released entry point declares a theorem with Mathlib’s RiemannHypothesis type. We read the target and the corresponding definition at the dependency’s pinned commit. The displayed generalised statements concern Dirichlet L-functions. That scope should travel with the claim; it should not silently become every possible generalisation of RH. These are source inspections, awaiting executed comparison of the complete proof environment. [S03] [S04]
What the Green Checks Establish
The verification project has successful CI jobs for two preliminary tasks: matching the submission to the published archive and compiling/exporting an independently stated challenge against Mathlib. We inspected the workflow, job steps and both logs. Neither job compiles the proposed solution’s Main module. [S05] [S07]
There is a useful trap here. The challenge file contains sorry placeholders because it states the questions that the solution must answer. The official checking procedure explicitly permits this. Their presence in that file is not evidence that the submitted solution is incomplete. The solution’s own proof dependencies must be checked separately. [S08] [S09]
The verifier’s notes still report the full solution build in progress. Reports of a source-wide text scan and resource requirements are attributed to that project; AkashicNET did not rerun the full scan or build. A truncated API tree was excluded from our source-count conclusions. [S06]
What Would Make the Claim Stronger?
A useful review keeps three proof checks distinct: the formal statement must express the intended mathematics, the proof must use only the permitted assumptions and the complete proof terms must be accepted by the checking kernels. Comparator is designed to compare a separately stated challenge with the submitted solution, including their definitions and axiom dependencies. A configuration file shows how a check is intended to run; its completed output supplies the result. [S08] [S09]
Here, the configured permitted assumptions are Lean’s standard propext, Quot.sound and Classical.choice. The author’s phrase about having no assumptions should therefore be read as a claim of no additional mathematical assumptions, awaiting the actual audit. A clean text search cannot replace that audit. [S07]
Mathematical understanding is another task. A faithful, successfully checked formal proof can establish a theorem even before people have a short explanation of its ideas. Conversely, an elegant explanation cannot stand in for a valid argument. Tao’s dated note directs readers towards ongoing verification and digestion, rather than giving a completed proof verdict. [S10]
Keep the Earlier Result in Its Own Frame
Anthropic’s August research post described an improvement in a lower bound for the fraction of zeta zeros satisfying RH. It explicitly said that work had not proved the hypothesis. Today’s released proof claim is a different event. Earlier related progress is relevant context, while each new argument needs its own checks. [S11]
The Millennium Prize has a separate process. Clay requires a qualifying publication, at least two years after that publication and general acceptance in the mathematics community before it considers a proposed solution. A proof announcement does not complete those conditions. Clay’s current problem page still says Unsolved; that label should not be used as a substitute for reviewing the new submission. [S12] [S13]
Wonder, Understanding and Responsibility
If the argument survives the required checks, the mathematical consequence would be profound. The result would concern the theorem actually proved, with its definitions and scope. Further claims about physics, encryption or consciousness would still require their own arguments and evidence.
For AkashicNET, the practical lesson is already useful: trace the headline to an inspectable object, check what each success signal measures and make uncertainty easy to find. Respect for the work includes careful credit, patient scrutiny and a willingness to correct the record.
What Would Change This Answer?
- A completed, reproducible proof check: inspect the exact source revision, statement comparison, final axiom dependencies and kernel verdicts before upgrading the reported status.
- A reproducible defect: state the failing theorem or checking step, the tested revision and the proposed repair. A build or resource failure alone does not refute the mathematics.
- An author correction or new release: preserve the earlier edition and repeat the affected checks.
- A clearer mathematical account: add an explanation while retaining the distinction between understanding, formal validity and prize acceptance.
Conclusion for this edition: a serious, inspectable claim merits scrutiny. The evidence reviewed here supports publication of this inquiry with its pending status; it does not support announcing a verified solution.
Full AkashicNEXUS Review
Maximum-depth manual appraisal. All 21 components considered; 18 applied and three not applicable. The same assistant researched, enhanced and checked this report. No independent report validation or general effectiveness estimate is claimed.
Material Claims
| ID | Claim | Judgement | Scope / Limit | Sources |
|---|
| C01 | A note on Alpöge’s website claims RH and a generalised result developed with Claude. | Supported as an attributed claim | This establishes what is being claimed, not that the argument is correct. | S02, S10 |
| C02 | The supplied post establishes a formal corporate announcement by Anthropic. | Unresolved / attribution narrowed | The primary announcement inspected is a personal-site note. No current corporate announcement was established by this review. | S01, S02, S11 |
| C03 | The released entry point names Mathlib’s RiemannHypothesis. | Supported by source inspection | The source target matches the standard definition on inspection. Transitive statement identity still needs executed comparison. | S03, S04 |
| C04 | The displayed GRH statements cover Dirichlet L-functions. | Supported by statement inspection | Do not extend the result to every L-function or to the expected Extended RH. | S03, S07 |
| C05 | The green CI checks validate the complete proposed proof. | Contradicted by the inspected workflow | The two jobs compare the archive and build/export the independent challenge. Neither builds Main. | S05, S07 |
| C06 | Challenge.lean containing sorry shows the submitted proof has a hole. | Contradicted as an inference | The challenge deliberately records unproved targets. The solution must discharge them separately. | S07, S08, S09 |
| C07 | The solution has passed its axiom audit and independent proof replay. | Unresolved | No completed result was present in the inspected verification record. AkashicNET did not execute these checks. | S06, S07 |
| C08 | The August partial-zero result and today’s proof claim are the same result. | Contradicted by date and scope | Retain both events separately. Earlier related progress neither proves nor disproves this submission. | S02, S11 |
| C09 | A proof claim immediately qualifies for the Millennium Prize. | Contradicted by the prize rules | Prize eligibility has its own qualifying-publication, elapsed-time and acceptance conditions. | S13 |
| C10 | This material establishes a quantum, spiritual or conscious AI mechanism. | Unsupported | The inspected objects concern a mathematical theorem and proof validation. They do not test subjective experience or metaphysical causation. | S02–S08 |
| C11 | A proof of RH by itself supplies a practical attack on encryption. | Unsupported by this material | No such algorithm or demonstrated attack is part of the inspected theorem targets. Implications require their own argument. | S03, S07 |
| C12 | A second assistant review makes the AkashicNET report independently validated. | Unsupported | This report was researched, revised and checked by the same assistant. Human owner review and independent external review remain separate. | S14 |
ENHANCE: Before and After
| ID | Before | After | Reason |
|---|
| E01 | Corporate shorthand in the starting post | A note on Alpöge’s website claims work with Claude | Preserves the verified source and the human collaborator. |
| E02 | An unrestricted use of GRH | GRH for the Dirichlet L-functions in the released statements | Stops a broader theorem being inferred from a narrower target. |
| E03 | A green badge treated as proof acceptance | Archive comparison and challenge build/export are the completed checks | Matches the summary to the actual jobs and logs. |
| E04 | Any occurrence of sorry treated as a failed submission | Challenge placeholders distinguished from the solution’s proof obligations | Avoids a false alarm while keeping the solution axiom audit pending. |
| E05 | A truncated API listing used as a complete inventory | Incomplete tree excluded from source-count claims | Keeps incomplete retrieval from becoming a misleading denominator. |
| E06 | An impressive claim treated as an established result | Claimed proof; full verification pending in the inspected record | Keeps the published conclusion aligned with available evidence. |
TEST and VALIDATOR
| ID | Check | Status | Observed Result |
|---|
| T01 | Source attribution | Checked | Personal-site claim retained; a corporate announcement was not established. |
| T02 | Theorem and dependency target | Checked · static scope | The RH target is present; the displayed GRH targets concern Dirichlet L-functions. |
| T03 | CI job scope | Checked | Archive comparison and challenge compilation/export succeeded. No Main build is in this workflow. |
| T04 | Legitimate challenge placeholders | Checked | Expected placeholders in the question file are not counted as a failed solution proof. |
| T05 | Wrong-scope claims | Checked · manual | Overextended GRH, conflated dates, immediate prize and unrelated implications rejected or qualified. |
| T06 | Full solution compilation | Not executed here | The verifier reports the build in progress. No local build pass is claimed. |
| T07 | Final axioms and kernel replay | Pending | No final axiom output or full Comparator verdict was available in that record. |
| T08 | NEXUS record contract | Passed · structural scope | Current checker returned valid: true with no structural errors. It does not establish source truth, mathematical validity, identity or independent review. |
| T09 | Article and infographic delivery | Passed · static scope | Source anchors, declared image dimensions, claim IDs and module coverage checked. Browser and deployment results are separate delivery checks. |
All Components
| Component | Edition | Scope | Output |
|---|
| AkashicMESSAGE | v0.4.0-beta.1 | Applied | Split the post into attributed report, proof claim, implication and publication request; preserved stable claim IDs. |
| AkashicINQUIRY | v0.3.0-beta.1 | Applied | DISCERN traced primary sources; DEEPEN checked code, CI scope, challenge placeholders and alternatives before synthesis. |
| AkashicASSESS | v0.3.0-beta.1 | Applied | Compared declaring a solution, withholding all coverage and publishing a bounded inquiry. Chose a dated inquiry with an explicit pending verdict. |
| AkashicENHANCE | v0.2.0-beta.1 | Applied | Six before/after changes retained, including the authority, GRH-scope and CI corrections. No increase in truth confidence inferred from a longer workflow. |
| AkashicSPIRIT | v0.3.0-draft.1 | Applied · reflective scope | Wonder can motivate inquiry and humility can guide attribution. Neither feeling supplies mathematical evidence. |
| AkashicABSTRACT | v0.1.0 | Applied | Summary preserves collaborator, claimed theorem scope, completed checks and missing proof verdict. |
| AkashicPRISM | Numbered edition unverified | Applied · manual draft scope | Reframed “did AI solve it?” into distinct questions about release, statement identity, proof validity, understanding and credit. |
| AkashicPYRAMID | Numbered edition unverified | Applied · layered reading | Foundation: source and definitions. Next: executed statement/axiom/kernel checks. Above: expert explanation and wider consequences. Missing steps remain visible. |
| AkashicONE | Numbered edition unverified | Not applicable | The linked private puzzle experiment is not required for this theorem review and was not accessed. |
| AkashicOMNI | v0.5.0 | Applied · manual v0.5.0 | All twelve published perspectives considered below; synthesis retains the same unresolved proof status. |
| AkashicNEURO | Numbered edition unverified | Not applicable | No neural, clinical or physiological measurements are present in the reviewed material. |
| AkashicTRACE | Numbered edition unverified | Applied | Original post, announcement, two pinned repositories, dependency pin, job IDs, access limits and retrieval timestamps retained. |
| AkashicTEST | Numbered edition unverified | Applied · bounded checks | Correct and misleading interpretations checked, including the legitimate sorry placeholders. Compilation and replay remain explicitly unexecuted. |
| AkashicFLOW | v0.2.0 | Applied | Source review, enhancement, checks and the new inquiry are linked. Owner review is a post-run handoff, not an invented publishing gate. |
| AkashicNEXUS | v0.1.0 | Applied | Coordinated one bounded manual route. All registered components were considered; no claim of autonomous engines or all-permutation execution. |
| AkashicSTATE | v0.1.0-draft.1 | Applied | Content edition v1.0.0 recorded. Scientific claim unresolved; owner and external reviews pending. No framework maturity ratings promoted. |
| AkashicFOOTPRINT | v0.1.0-alpha.1 | Applied · measurement limits | No complete labour, billed-usage, energy or carbon measurements collected. No efficiency or cost saving claimed; local full-proof build not attempted. |
| AkashicTIER | v0.1.0-draft.1 | Not applicable | No package, entitlement or commercial offer was requested. |
| AkashicAUDIT | v0.1.0-draft.3 | Applied · self-review | Every material claim has a source and a scoped judgement. Missing evidence remains unresolved, not false. |
| AkashicLEDGER | v0.1.0-draft.2 | Applied | Starting report, six refinements, check outcomes and reopening triggers preserved in this dated record. |
| AkashicVALIDATOR | v0.1.1-alpha.1 | Applied · structural profile | Current component v0.1.1-alpha.1; checker and contract v0.1.2. Structural execution result is added after running it. |
Twelve OMNI Perspectives
| Perspective | Application |
|---|
| AWAKEN | Let the significance invite attention while keeping the theorem verdict open. |
| HIERATIC | Read the language of beauty and revelation as the author’s expression, separately from the formal argument. |
| HOMESENSE · Phase 8 | Ask what a person can inspect, explain and rely on when the argument is too large to follow unaided. |
| ADAPT | Use independently stated targets and checkable artifacts to adapt review practice to generated proofs. |
| REGENERATE | Treat maintained explanations and reusable checking methods as potential public benefits, not already measured outcomes. |
| TRANSCEND | Explore a change in how mathematics is produced without abandoning criteria for validity. |
| #METAD | Notice how a respected name, a green badge or the word Lean can change confidence before the proof is checked. |
| ACTC | Separate the human’s account of collaboration from evidence assigning causal credit to specific model actions. |
| MultidimensionalCUT · Past | Keep the earlier partial result and longstanding mathematical tools in their own dated source trail. |
| MultidimensionalCUT · Present | The actionable objects are the released targets, dependency pins and current verification outputs. |
| MultidimensionalCUT · Future | A pass, a reproducible flaw or an inconclusive resource limit leads to a different next step. |
| UMASC | Distinguish individual understanding, machine checking and community acceptance as different scales of assessment. |
Post-Run Human Review
Review by the curator is pending. The curator is invited to assess accuracy, sources, gaps and presentation. Answers are visible. Independent external review of this appraisal: Pending. External mathematical checking is a different process.
Framework readiness ratings are retained. No complete labour, billed usage, energy or carbon measurements were collected. No savings are claimed.
Run NEXUS-RIEMANN-2026-11-20-001 · v1.0.0 · NEXUS v0.1.0 · OMNI v0.5.0 · Contract and checker v0.1.2.
Source mirror: 7cbf7a0c32fca1badf05931879a37c3f8a2b03b4
Verification revision: 1f2beda145f7b7a9f6f6f6640a60f2e44920b7ba
Mathlib pin: 0df444a360eaa60ab8c11dca51a86af692955474
The Inquiry in One Image
AkashicNET editorial infographic · Evidence snapshot 20 Nov 2026, 19:40 CET. Schematic status summary. The 29 is a personal design motif.