Claude searched protein databases, designed DNA constructs, reduced purification from about three days to about 90 minutes, and processed cryo-EM data into an initial model in 34 hours. The result is a particle made of 96 copies of a bacteriophage protein, with about 25% more internal volume than the mammalian version. Its biological function is unknown; this is a preprint, and Claude made errors that humans corrected.
Technology: Claude
Family Claude: profile · models page
Publication: Preprint · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: direct test
- Tested: The 3D structure was determined by cryo-EM; cryo-EM confirmed the vault shape for four of seven particle-forming proteins, and fifteen altered protein versions were tested, with a reviewer checking the structure.
- Still unknown: The biological function is unknown, vaults were made only in E. coli, and an unidentified pocket molecule resembling coenzyme A is unconfirmed.
- AI role: Claude searched protein databases, designed DNA constructs, optimized purification, and ran RELION cryo-EM processing to an initial model in 34 hours; it also made errors that humans corrected.
Evidence sources: https://www.anthropic.com/research/investigating-vault-particles-with-claude
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
Anthropic reports early results with six partners: a CAR T-cell receptor design about 70% shorter that enabled immune cells to produce about 10 times more receptors; about 35,000 antibody designs in one month, yielding 20 candidates; and a document review that found all 11 human-identified issues plus seven more in 89% less time. The work is mostly preclinical and reported by Anthropic.
Technology: Claude
Family Claude: profile · models page
Publication: Announcement only · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Announcement only
- Test type: direct test
- Tested: Partner lab tests and a document review reported by Anthropic: a CAR T receptor design was tested in immune cells, antibody designs were screened for IgE binding, and a review of 1,061 pages was compared with 11 human-identified issues.
- Still unknown: The work is mostly preclinical with no human tests; partner reports give limited methodological detail, and the page does not say whether the seven extra flags were confirmed.
- AI role: Claude proposed molecular and receptor designs, generated about 35,000 antibody designs, reviewed manufacturing documents, and drafted regulatory documents.
Evidence sources: https://www.anthropic.com/research/early-progress-drug-development
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: not linked in this catalog; this does not mean no paper exists.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
The paper reports a computational mechanism for perrhenate-catalysed deoxydehydration: proton-assisted water loss enables ring closure and alcohol reduction. Under human-defined hypotheses, a Claude Code agent located transition states, which the authors validated. Coupled-cluster-anchored microkinetic modeling predicted a rate increase of about eight orders of magnitude over the best proton-free pathway. At matched conversion conditions, modeled barriers were consistent with the 31.5–34.0 kcal/mol window inferred from previously reported experimental turnover; this was not a new wet-lab demonstration.
Technology: Claude Code LLM agent for human-directed reaction-network and transition-state search, combined with electronic-structure calculations and microkinetic modeling
Family Claude: profile · models page
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: Agent-located transition states verified by the authors; computed energetics, reaction-network and microkinetic consistency checks; comparisons to existing turnover-derived barrier estimates.
- Still unknown: Main paper reviewed; supplementary files were not read. The mechanism and rate changes are predictions compared with prior experimental kinetics, not a new experimental confirmation.
- AI role: Claude Code carried out calculation/search/triage workflows for author-specified mechanistic hypotheses; authors validated the located stationary points.
Evidence sources: https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A%2210.1039%2Fd6sc06865k%22&resultType=core&format=json · https://www.ebi.ac.uk/europepmc/webservices/rest/PMC13720772/fullTextXML
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
In a mostly autonomous 64-hour campaign, Claude designed minibinders against five neurotoxins. External labs tested 211 designs, 17 of which met the criteria; the best (Cn2-B1 and Cn2-B2) had affinities in the low nanomolar range. Only recombinant toxins were tested, with no neutralization tests in cells or animals; Anthropic does not claim to have produced an antivenom.
Technology: Claude
Family Claude: profile · models page
Publication: Announcement only · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Announcement only
- Test type: direct test
- Tested: External labs tested 211 designs by surface plasmon resonance: Adaptyv Bio in cell-free expression and Twist Bioscience in human kidney cells; a binder needed KD of 1 µM or tighter in every replicate at both labs and no CLEC12A signal, giving 17 binders. The two collaborating external laboratories tested candidates from the discovery project; this establishes direct testing rather than a separately established independent follow-up replication.
- Still unknown: Only recombinant toxins were tested, with no neutralization in cells or animals; the authors do not know whether the binders neutralize venom and do not claim an antivenom; multitoxin designs failed.
- AI role: Claude selected toxin structures and hotspots, generated 318,484 designs, filtered and co-folded them, ran about 24 hours autonomously, then produced 180 selected designs after human follow-ups.
Evidence sources: https://www.anthropic.com/research/claude-scorpion-toxin-binder-design
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: not linked in this catalog; this does not mean no paper exists.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
Claude designed circuits for eight basic operations (chemistry, optimization, physical simulation, and number theory), assessed by an automated verifier. In several cases they outperform the best published circuits; for example, a point addition used 17 times fewer gates at the cost of 12% more qubits. Key identities were checked with simulation and Lean; a memory shortcut in the decoder fails about once in 10,000 times.
Technology: Claude
Family Claude: profile · models page
Publication: Preprint · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: retrospective or benchmark
- Tested: An automated evaluator checked each circuit for correctness and scored qubits times gates; results were compared with published designs, divsteps was checked by simulation and Lean, and decoder memory was measured over 403 million simulated decodings.
- Still unknown: The decoder memory shortcut fails about once in 10,000 runs; gains apply to individual primitives, not a full fault-tolerant machine; Claude was conservative and needed prompts to broaden its search.
- AI role: Claude wrote circuits as Python code and revised them in a closed loop with evaluator feedback, with no human input between rounds, and reconstructed branch information to reduce memory.
Evidence sources: https://www.anthropic.com/research/designing-efficient-quantum-circuits
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
Alman and Vassilevska Williams present deterministic algorithms for 3SUM on polynomial-size integers in O(n^1.9992) time and all-pairs shortest paths on directed graphs with polynomially bounded integer weights in O(n^2.9995) time, the first polynomial improvements over the textbook exponents. The single underlying technique computes selected entries of thin matrix products more efficiently, and known reductions yield the main algorithmic consequences. The paper says an internal Anthropic Claude research model discovered the algorithm autonomously; the human authors then understood, simplified, strengthened and extended it. One paper-level theoretical-computer-science result is counted.
Technology: Internal Anthropic Claude research model (specific model version undisclosed)
Family Claude: profile · models page
Publication: Preprint · Test context: prospective or independent check. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: prospective or independent check
- Tested: Alman and Vassilevska Williams present deterministic algorithms for 3SUM on polynomial-size integers in O(n^1.9992) time and all-pairs shortest paths on directed graphs with polynomially bounded integer weights in O(n^2.9995) time, the first polynomial improvements over the textbook exponents.
- Still unknown: This is an arXiv preprint as of the checked record. The stated time bounds are for polynomially bounded integer inputs; real-valued variants use Las Vegas randomization and different bounds. Lean certification covers the explicitly enumerated theorems, not every statement or presentation detail in the 76-page paper. The authors take responsibility for the paper and developed important extensions/consequences after the initial AI-discovered algorithm.
- AI role: The paper states that Claude was tasked with checking and improving cryptographic constructions but instead developed the new thin-matrix-product algorithm, first for average-case and then worst-case problems, in a 16-million-output-token session with no human input. The authors subsequently worked to understand, simplify, strengthen and extend the algorithm. Anthropic later used an internal model to Lean-certify specified main results.
Evidence sources: https://arxiv.org/abs/2610.06783
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 4/5 · Validation 4/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
Zhao Song and Xinzhi Zhang prove a universal, dimension-independent bound for the Kannan–Lovász–Simonovits constant of isotropic log-concave probability measures, resolving the KLS conjecture. Their proof repeatedly refines bounds on polynomial variance and curvature; the authors report that AI tools helped explore more than 100 approaches, obtain the earlier log-star bound, and develop and verify the final proof. Independent researchers also posted proofs, and the paper’s streamlined proof is being formalized in Lean. The 1 October preprint gave a slowly dimension-dependent bound; the 4 October version established the dimension-independent theorem described here.
Technology: GPT-6 Astra; GPT-5.6 Sol; Claude Fable 5; Fable (as named in the paper’s AI disclosure)
Family Claude: profile · models page
Publication: Preprint · Test context: prospective or independent check. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: prospective or independent check
- Tested: Zhao Song and Xinzhi Zhang prove a universal, dimension-independent bound for the Kannan–Lovász–Simonovits constant of isotropic log-concave probability measures, resolving the KLS conjecture.
- Still unknown: The proof is a recent arXiv preprint, not yet journal peer-reviewed in the source reviewed. Independent proofs were posted, and the Microsoft Research author account reports later Lean formalization of a streamlined proof; the paper version and posted revisions should be checked as the work develops. This entry counts the theorem result once, not the intermediate log-star bound separately.
- AI role: The authors state that they explored candidate approaches collaboratively with AI tools; the tools helped obtain the earlier log-star bound and develop detailed proof steps, while the researchers selected the direction and completed the repeated refinement argument. The authors state they checked the proofs with AI tools and take responsibility for correctness.
Evidence sources: https://arxiv.org/pdf/2610.01447v2 · https://arxiv.org/abs/2610.01447 · https://www.microsoft.com/en-us/research/blog/solving-a-31-year-old-grand-challenge-in-mathematics-the-kls-conjecture/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 4/5 · Validation 4/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
About 950 Claude agents searched a DNA-sequence database for 21 hours (210 million tokens) and flagged an uncharacterized enzyme system, found mostly in bacteriophages, with a DNA repeat array similar to CRISPR. Scientists performed the laboratory work; the function remains unknown, and the work is a preprint, not a peer-reviewed publication.
Technology: Claude (agents)
Family Claude: profile · models page
Publication: Preprint · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: direct test
- Tested: Human scientists expressed the protein in laboratory strains and characterized it biochemically and structurally; early experiments showed the array is expressed as distinct short RNAs.
- Still unknown: The primary function of the enzyme system is unknown; findings are preliminary, and most Claude-proposed candidates are eliminated in follow-up review.
- AI role: About 950 agents used 210 million tokens over 21 hours to search DNA sequences, narrowing about 3,500 candidate systems to 20; one agent noticed the repeat array and filed a report.
Evidence sources: https://anthropic.com/news/claude-discovers-novel-enzyme-system
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
In just under four weeks, Claude optimized over 30 models for structure, protein design, and genomics. They were on average four times faster with a small accuracy loss, and nearly twice as fast with identical outputs. It also created a low-memory mode for systems over 10,000 tokens on a single GPU node. Design results are in silico only, and comparisons are internal to Anthropic.
Technology: Claude (in Claude Science)
Family Claude: profile · models page
Publication: Announcement only · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Announcement only
- Test type: retrospective or benchmark
- Tested: Accelerated models were checked against their downstream tasks (for example structure prediction) and FlashPairformer kernels were compared with cuEquivariance (2.7-2.9x faster on triangle attention); design results are in silico with internal comparisons.
- Still unknown: Design results have no wet-lab validation, large viral capsid predictions were incorrect, ColabFold kernels were not benchmarked, and the model is not reproducible.
- AI role: Claude wrote custom FlashPairformer kernels, optimized more than 30 models, built a low-memory Big mode, and ran autonomous protein binder design.
Evidence sources: https://www.anthropic.com/research/claude-uplifts-biomolecular-modeling
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: not linked in this catalog; this does not mean no paper exists.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
A proof of Fermat's Last Theorem fully machine-verified in Lean: Claude produced most of it independently over 11 days, with about 13 million lines and 29,500 intermediate theorems, and a comparator confirms that its statement matches Mathlib's. It follows a simplified version of Wiles's proof, uses Mathlib and parts of an Imperial College project, and is probably longer than necessary.
Technology: Claude, Lean
Family Claude: profile · models page
Publication: Announcement only · Test context: prospective or independent check. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Announcement only
- Test type: prospective or independent check
- Tested: The proof was checked by Lean using only standard axioms; a comparator tool confirmed the theorem statement matches Mathlib's, and mathematician Kevin Buzzard reviewed it.
- Still unknown: The proof is about 13 million lines and likely much longer than needed; it follows a simplified version of Wiles's proof and is presented as verification, not new mathematics.
- AI role: Dozens of agents worked for 11 days on the Prove2Me platform, producing about 30,300 computer-verified theorems; humans gave occasional high-level instructions.
Evidence sources: https://www.anthropic.com/research/formalizing-fermats-last-theorem
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: not linked in this catalog; this does not mean no paper exists.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 4/5 · Validation 4/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
After an initial prompt, Claude selected binding sites, used design and folding models, and chose candidates, with success on 14 of 15 targets. Of 1,320 designs, 354 bound, for rates of 22–35% versus a typical 10–15%. None of the 90 designs for the maltose-binding protein worked. External labs ran the tests.
Technology: Claude (Mythos Preview and Opus 4.8)
Family Claude: profile · models page
Publication: Announcement only · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Announcement only
- Test type: direct test
- Tested: External labs (Adaptyv Bio and Twist Bioscience) produced and tested the designs: 354 of 1,320 designs bound, with hit rates of 22-35% versus a typical 10-15%; none of 90 designs for maltose-binding protein bound. The collaborating external laboratories performed candidate production and binding assays within the discovery project; separate independent follow-up replication is not established by this account.
- Still unknown: Anthropic plans further characterization to confirm hit rates and affinities; BBF-14 affinities are modest, GDF-8 data were inconclusive and excluded, and the TNFa result is unexplained.
- AI role: Claude chose binding sites, generated candidate structures and sequences by orchestrating structure, sequence and co-folding models, ran in silico optimization and chose candidates.
Evidence sources: https://www.anthropic.com/research/Claude-accelerates-protein-design
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: not linked in this catalog; this does not mean no paper exists.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
Giorgio Parisi and Francesco Zamponi provide an analytic proof that the critical exponents a and b in the full replica-symmetry-breaking scaling solution for infinite-dimensional dense hard spheres obey a+b=1. The identity links gap, contact-force and overlap scaling relations and had previously been known numerically but not analytically. The authors report that Claude Opus 4.7 derived the core proof idea with minimal supervision, with Sonnet 4.6 helping refine it; the authors found and corrected errors in an early draft, then verified and edited the final proof.
Technology: Anthropic Claude Opus 4.7 and Claude Sonnet 4.6
Family Claude: profile · models page
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: Giorgio Parisi and Francesco Zamponi provide an analytic proof that the critical exponents a and b in the full replica-symmetry-breaking scaling solution for infinite-dimensional dense hard spheres obey a+b=1.
- Still unknown: The theorem is scoped to the fullRSB scaling equations for the infinite-dimensional dense hard-sphere jamming solution; it is not a general proof about every finite-dimensional jamming system. The result is peer-reviewed and published, but proof validation and correction are attributed to the human authors.
- AI role: After numerical investigation, Opus generated the analytic proof strategy and proof draft for the previously unproven scaling identity; Sonnet refined minor steps. The authors report 40 prompts, corrected early inconsistencies and verified/edited the published proof.
Evidence sources: https://doi.org/10.1088/1742-5468/ae7bd7
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
Physicist Matthew Schwartz (Harvard) guided Claude Opus 4.5 through a resummation calculation for e+e− collisions, producing a paper with a new factorization theorem and testable predictions against data in about two weeks. Schwartz checked every step; the model claimed checks it had not done and made errors.
Technology: Claude Opus 4.5
Family Claude: profile · models page
Publication: Preprint · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: retrospective or benchmark
- Tested: Results are validated against EVENT2 Monte Carlo simulations, with analytic curves matching the histograms; the paper does not describe a comparison with experimental data, and the author checked each step, with GPT and Gemini cross-checking the written calculations.
- Still unknown: The paper does not describe comparison with experimental data (the author mentions a follow-up project for this); the author states that LLMs cannot yet do original theoretical physics autonomously and that the work needed expert evaluation.
- AI role: Claude Opus 4.5, run through Claude Code, performed the calculations including the factorization derivation, numerical analysis, simulations and manuscript preparation, under physicist supervision.
Evidence sources: https://www.anthropic.com/research/vibe-physics · https://arxiv.org/abs/2601.02484
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
A Claude Code agent pipeline analyzed archived 1994 ALEPH hadronic-Z data and reported the first primary Lund jet-plane density measurement in e⁺e⁻ collisions. Its unfolded 57-bin fiducial result shows the expected triangular QCD radiation structure and a shallow low-transverse-momentum turnover; the integrated mean primary-emission count is 4.751 ± 0.224 per hemisphere. This is one new measurement, not a claim of discovering new particles or physics beyond QCD.
Technology: Claude Code agents orchestrated by the Just Furnish Context (JFC) framework
Family Claude: profile · models page
Publication: Preprint · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: retrospective or benchmark
- Tested: A Claude Code agent pipeline analyzed archived 1994 ALEPH hadronic-Z data and reported the first primary Lund jet-plane density measurement in e⁺e⁻ collisions.
- Still unknown: This is an arXiv preprint, and the authors explicitly qualify the novel measurement as subject to independent verification. It uses 1994 ALEPH data and one available full-simulation sample; the corrected density is prior/model dependent, with a reported 6.0% density-weighted total uncertainty and 4.7% uncertainty on the integrated count. The shallow residual shape is within current generator-theory uncertainty, so the paper frames it as a benchmark rather than evidence excluding generators.
- AI role: Agents planned and executed the event selection, jet declustering, unfolding, uncertainty propagation, statistical checks and report drafting using archived ALEPH data and literature retrieval. The paper attributes the new Lund-plane measurement to an autonomous AI-agent analysis pipeline; human authors supplied the broad task and review framework.
Evidence sources: https://arxiv.org/html/2603.20179v3
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
A collaborative LLM-plus-symbolic-computation study proved that every n×n Latin square with n ≡ 1 (mod 3) has imbalance at least 4n(n−1)/9. It introduced near-perfect permutations that attain the bound and computationally found independently verified matching examples for every such order from 4 through 52. The universal lower-bound theorem was formally checked in Lean 4.
Technology: Claude Opus 4.5 LLM agent with SageMath, Rust exhaustive solver, simulated annealing and Lean 4 verification tools
Family Claude: profile · models page
Publication: Preprint · Test context: prospective or independent check. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: prospective or independent check
- Tested: A collaborative LLM-plus-symbolic-computation study proved that every n×n Latin square with n ≡ 1 (mod 3) has imbalance at least 4n(n−1)/9.
- Still unknown: This is an arXiv preprint (v2 dated 2026-08-13), not yet peer-reviewed in the sources checked. The universal proven theorem is a lower bound for n ≡ 1 mod 3. Matching constructions, and hence demonstrated tightness, are computationally established only for n=4,7,…,52; the paper does not prove that the bound is attained for every larger order. Lean 4 formalizes the lower-bound inequality, not the finite construction search or universal tightness.
- AI role: The agent identified an all-even row-distance parity pattern in numerical search data, formulated and helped prove the lower bound, and used model review to catch proof errors. SageMath and a Rust solver supported symbolic analysis and enumeration; simulated annealing found matching constructions through n=52. Human strategic direction supplied the decisive pivot to the minimum-positive-imbalance question. Lean 4 formally verifies the lower-bound theorem itself.
Evidence sources: https://arxiv.org/html/2603.08322v2
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 4/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
Knuth's note reports that Claude Opus 4.6 found an explicit decomposition of the arcs of the m^3-vertex digraph into three directed Hamiltonian cycles for odd m, an open problem Knuth posed. Knuth's proof of the construction was reportedly formally verified in Lean by Kim Morrison. Caveat: sources disagree on the even case, which an independent arXiv paper (2603.24708) and a GPT-5.4 Pro proof treat differently.
Technology: Claude Opus 4.6
Family Claude: profile · models page
Publication: Not annotated in this catalog · Test context: Not annotated in this catalog. Catalog categories, not an evidence grade.
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: other. Paper: not linked in this catalog; this does not mean no paper exists.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 4/5 · Novelty 5/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
Anthropic describes three cases: Biomni (Stanford) completed a GWAS analysis in 20 minutes in an initial trial and suggested new transcription factors in embryonic development; MozzareLLM (MIT and Whitehead) interprets gene clusters from CRISPR screens; a third lab is still testing. This is an Anthropic account without independent evaluation; some figures are estimates.
Technology: Claude (Biomni, MozzareLLM)
Family Claude: profile · models page
Publication: Announcement only · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Announcement only
- Test type: retrospective or benchmark
- Tested: Biomni's case studies (GWAS analysis, cloning design blind-evaluated against a postdoc, embryonic-tissue cell analysis) were checked by the Biomni team, citing a bioRxiv preprint; MozzareLLM was assessed qualitatively by its lab, with no quantitative benchmark.
- Still unknown: The article says the validation is not independent, some figures are estimates (such as the three-week human comparison), and Biomni cannot yet do everything out of the box.
- AI role: A Claude-powered agent (Biomni) analysed GWAS and embryonic cell data and suggested transcription factors; a Claude-powered system (MozzareLLM) interprets gene clusters from CRISPR screens.
Evidence sources: https://www.anthropic.com/news/accelerating-scientific-research
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: not linked in this catalog; this does not mean no paper exists.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.