{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:VQMPFD7EHMUMO6FVYUPLBU6POS","short_pith_number":"pith:VQMPFD7E","canonical_record":{"source":{"id":"2605.07919","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CV","submitted_at":"2026-05-08T15:55:30Z","cross_cats_sorted":[],"title_canon_sha256":"53ac9af92f70e8a5751c4d6922585835fb90690dc00399e153d13ad726421550","abstract_canon_sha256":"133a9adea01e80b03ff81c4967815dba5865719a1437782d91a280f73ba236a3"},"schema_version":"1.0"},"canonical_sha256":"ac18f28fe43b28c778b5c51eb0d3cf74b15b8cf4701039e8417b656176d45bf0","source":{"kind":"arxiv","id":"2605.07919","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.07919","created_at":"2026-05-25T02:02:16Z"},{"alias_kind":"arxiv_version","alias_value":"2605.07919v2","created_at":"2026-05-25T02:02:16Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.07919","created_at":"2026-05-25T02:02:16Z"},{"alias_kind":"pith_short_12","alias_value":"VQMPFD7EHMUM","created_at":"2026-05-25T02:02:16Z"},{"alias_kind":"pith_short_16","alias_value":"VQMPFD7EHMUMO6FV","created_at":"2026-05-25T02:02:16Z"},{"alias_kind":"pith_short_8","alias_value":"VQMPFD7E","created_at":"2026-05-25T02:02:16Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:VQMPFD7EHMUMO6FVYUPLBU6POS","target":"record","payload":{"canonical_record":{"source":{"id":"2605.07919","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CV","submitted_at":"2026-05-08T15:55:30Z","cross_cats_sorted":[],"title_canon_sha256":"53ac9af92f70e8a5751c4d6922585835fb90690dc00399e153d13ad726421550","abstract_canon_sha256":"133a9adea01e80b03ff81c4967815dba5865719a1437782d91a280f73ba236a3"},"schema_version":"1.0"},"canonical_sha256":"ac18f28fe43b28c778b5c51eb0d3cf74b15b8cf4701039e8417b656176d45bf0","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-25T02:02:16.509641Z","signature_b64":"vAlHsULuIJSPBomTOcKUcnXkUxCkUfgsQcWyKLCbEvTHKByTIlHvd0vHkhGqKT0K8tc9Usamc06CS1fzRGrVDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ac18f28fe43b28c778b5c51eb0d3cf74b15b8cf4701039e8417b656176d45bf0","last_reissued_at":"2026-05-25T02:02:16.508920Z","signature_status":"signed_v1","first_computed_at":"2026-05-25T02:02:16.508920Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.07919","source_version":2,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-25T02:02:16Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"fybMlV7Z1qRSGQ/7vxQAkAsdSfHxRqFjDO4u85qgieEHlI3qioEsei/tZnkyfDmlN3Sk9aAe0HvEYBx4axnlCA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-02T10:20:35.912832Z"},"content_sha256":"d987f550ae76bdb920390ab8c740391d590ce01f56262d4c9ee595600804d93c","schema_version":"1.0","event_id":"sha256:d987f550ae76bdb920390ab8c740391d590ce01f56262d4c9ee595600804d93c"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:VQMPFD7EHMUMO6FVYUPLBU6POS","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"MedVIGIL: Evaluating Trustworthy Medical VLMs Under Broken Visual Evidence","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"Medical vision-language models fail to refuse answers when visual evidence is broken, trailing radiologists by 14 points on a new composite score.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Hanqi Jiang, Haozhen Gong, Hui Ren, Hyeokjae Kwon, Jinglei Lv, Junhao Chen, Lifeng Chen, Lin Zhao, Mingyu Kang, Quanzheng Li, Ruiyu Yan, Tianming Liu, Weihang You, Xiang Li, Yi Pan","submitted_at":"2026-05-08T15:55:30Z","abstract_excerpt":"Medical vision--language models (VLMs) are usually evaluated on intact image--question pairs, but trustworthy clinical use requires a stronger property: a model must recognise when the evidential basis for an answer has failed. We study this through silent failures under perturbed evidence, where a vision-required medical question is paired with a false premise, wording perturbation, knowledge-only rewrite, or ROI-corrupted image, yet the model returns a fluent non-refusal answer. We introduce medvigil, a 300-case evaluation suite drawn from four public medical VQA sources, supervised end to e"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"The independent radiologist scores MCS 83.3 at silent-failure rate 5.8%, leaving a 14.1-point composite headroom above the strongest audited model (Claude Opus 4.7 at 69.2).","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the four chosen perturbation types and the 300 clinician-authored cases sufficiently represent the range of broken visual evidence that occurs in real clinical practice.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"MedVIGIL introduces a clinician-supervised benchmark showing medical VLMs frequently give fluent answers on broken visual evidence, with top models 14 points below human radiologists on the composite score.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Medical vision-language models fail to refuse answers when visual evidence is broken, trailing radiologists by 14 points on a new composite score.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"5abf56945b417f28f53c287fd507a3ccfd3d0c308bedf4408748d780b5d4b2a2"},"source":{"id":"2605.07919","kind":"arxiv","version":2},"verdict":{"id":"40477877-05fe-4735-b79f-76bbcd099224","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-11T03:40:45.192215Z","strongest_claim":"The independent radiologist scores MCS 83.3 at silent-failure rate 5.8%, leaving a 14.1-point composite headroom above the strongest audited model (Claude Opus 4.7 at 69.2).","one_line_summary":"MedVIGIL introduces a clinician-supervised benchmark showing medical VLMs frequently give fluent answers on broken visual evidence, with top models 14 points below human radiologists on the composite score.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the four chosen perturbation types and the 300 clinician-authored cases sufficiently represent the range of broken visual evidence that occurs in real clinical practice.","pith_extraction_headline":"Medical vision-language models fail to refuse answers when visual evidence is broken, trailing radiologists by 14 points on a new composite score."},"integrity":{"clean":false,"summary":{"advisory":2,"critical":0,"by_detector":{"doi_compliance":{"total":2,"advisory":2,"critical":0,"informational":0}},"informational":0},"endpoint":"/pith/2605.07919/integrity.json","findings":[{"note":"DOI in the printed bibliography is fragmented by whitespace or line breaks. A longer candidate (10.1109/TBME.2021.3117407.URLhttps://pmc.ncbi.nlm.nih.gov/articles/PMC9011180/.Dan) was visible in the surrounding text but could not be confirmed against doi.org as printed.","detector":"doi_compliance","severity":"advisory","ref_index":10,"audited_at":"2026-05-19T11:24:20.741484Z","detected_doi":"10.1109/TBME.2021.3117407.URLhttps://pmc.ncbi.nlm.nih.gov/articles/PMC9011180/.Dan","finding_type":"recoverable_identifier","verdict_class":"incontrovertible","detected_arxiv_id":null},{"note":"DOI in the printed bibliography is fragmented by whitespace or line breaks. A longer candidate (10.18653/v1/2023.emnlp-main.20.URLhttps://aclanthology.org/2023.emnlp-main.20/.Bo) was visible in the surrounding text but could not be confirmed against doi.org as printed.","detector":"doi_compliance","severity":"advisory","ref_index":17,"audited_at":"2026-05-19T11:24:20.741484Z","detected_doi":"10.18653/v1/2023.emnlp-main.20.URLhttps://aclanthology.org/2023.emnlp-main.20/.Bo","finding_type":"recoverable_identifier","verdict_class":"incontrovertible","detected_arxiv_id":null}],"available":true,"detectors_run":[{"name":"claim_evidence","ran_at":"2026-05-20T10:02:13.555939Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-20T04:46:22.489233Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T15:31:18.310769Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T11:24:20.741484Z","status":"completed","version":"1.0.0","findings_count":2}],"snapshot_sha256":"4abdae5894b4b9e13c60c8601c856f602b2de3eb7fdd7894a72fce302b49f1cc"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"40477877-05fe-4735-b79f-76bbcd099224"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-25T02:02:16Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"CejPGsp3sH5r3Aev4aJBw+gS++K4w0MfAcPbTOGOEEw+U4TKkaviYgv/dyMtvK4Ooeztz/Dxaunw6kNlI4yrBQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-02T10:20:35.913410Z"},"content_sha256":"c3d4bac97121078b6e87557fb6ab519e093feb65cc4ce1c771e857c65c0ac20c","schema_version":"1.0","event_id":"sha256:c3d4bac97121078b6e87557fb6ab519e093feb65cc4ce1c771e857c65c0ac20c"},{"event_type":"integrity_finding","subject_pith_number":"pith:2026:VQMPFD7EHMUMO6FVYUPLBU6POS","target":"integrity","payload":{"note":"DOI in the printed bibliography is fragmented by whitespace or line breaks. A longer candidate (10.18653/v1/2023.emnlp-main.20.URLhttps://aclanthology.org/2023.emnlp-main.20/.Bo) was visible in the surrounding text but could not be confirmed against doi.org as printed.","snippet":"URL https://arxiv.org/abs/2510.10965. Yifan Li, Yifan Du, Kun Zhou, Jinpeng Wang, Xin Zhao, and Ji-Rong Wen. Evaluating object hallucination in large vision-language models. InProceedings of the 2023 Conference on Empir- ical Methods in Nat","arxiv_id":"2605.07919","detector":"doi_compliance","evidence":{"ref_index":17,"verdict_class":"incontrovertible","resolved_title":null,"printed_excerpt":"URL https://arxiv.org/abs/2510.10965. Yifan Li, Yifan Du, Kun Zhou, Jinpeng Wang, Xin Zhao, and Ji-Rong Wen. Evaluating object hallucination in large vision-language models. InProceedings of the 2023 Conference on Empir- ical Methods in Nat","reconstructed_doi":"10.18653/v1/2023.emnlp-main.20.URLhttps://aclanthology.org/2023.emnlp-main.20/.Bo"},"severity":"advisory","ref_index":17,"audited_at":"2026-05-19T11:24:20.741484Z","event_type":"pith.integrity.v1","detected_doi":"10.18653/v1/2023.emnlp-main.20.URLhttps://aclanthology.org/2023.emnlp-main.20/.Bo","detector_url":"https://pith.science/pith-integrity-protocol#doi_compliance","external_url":null,"finding_type":"recoverable_identifier","evidence_hash":"9ec76dbec851fce9a6de1047e81fc5a401173f20a9666e46d704617d58384de8","paper_version":1,"verdict_class":"incontrovertible","resolved_title":null,"detector_version":"1.0.0","detected_arxiv_id":null,"integrity_event_id":1030,"payload_sha256":"ed482f6cf19f54defa60a7f5b9a9e7da3c001a7031aa99ced41bba6d46180b87","signature_b64":"JZJ+oMPI7Wk3+sodLUamw1AE+WcVQsQQ9TbVt8Lrsk7LmylJSkazkayieb+bxn/OTS+qvkCAvoy6YLLUl76XCA==","signing_key_id":"pith-v1-2026-05"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-19T11:27:01Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"w6wJ0gyX4Ngogss8CnCCbvmlOFZrcVqS0Opr+z4yOnduyYdNKA6w88mydaVKr5TXwf7ai9R/7vO1cIKPabOMDA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-02T10:20:35.914286Z"},"content_sha256":"c7932fd0ad5b6493465116afc44e150f7ff7f89cde51d18a76123ed4553b00f1","schema_version":"1.0","event_id":"sha256:c7932fd0ad5b6493465116afc44e150f7ff7f89cde51d18a76123ed4553b00f1"},{"event_type":"integrity_finding","subject_pith_number":"pith:2026:VQMPFD7EHMUMO6FVYUPLBU6POS","target":"integrity","payload":{"note":"DOI in the printed bibliography is fragmented by whitespace or line breaks. A longer candidate (10.1109/TBME.2021.3117407.URLhttps://pmc.ncbi.nlm.nih.gov/articles/PMC9011180/.Dan) was visible in the surrounding text but could not be confirmed against doi.org as printed.","snippet":"doi: 10.1109/TBME.2021. 3117407. URLhttps://pmc.ncbi.nlm.nih.gov/articles/PMC9011180/. Dan Hendrycks and Thomas Dietterich. Benchmarking neural network robustness to common corruptions and perturbations. InInternational Conference on Learni","arxiv_id":"2605.07919","detector":"doi_compliance","evidence":{"ref_index":10,"verdict_class":"incontrovertible","resolved_title":null,"printed_excerpt":"doi: 10.1109/TBME.2021. 3117407. URLhttps://pmc.ncbi.nlm.nih.gov/articles/PMC9011180/. Dan Hendrycks and Thomas Dietterich. Benchmarking neural network robustness to common corruptions and perturbations. InInternational Conference on Learni","reconstructed_doi":"10.1109/TBME.2021.3117407.URLhttps://pmc.ncbi.nlm.nih.gov/articles/PMC9011180/.Dan"},"severity":"advisory","ref_index":10,"audited_at":"2026-05-19T11:24:20.741484Z","event_type":"pith.integrity.v1","detected_doi":"10.1109/TBME.2021.3117407.URLhttps://pmc.ncbi.nlm.nih.gov/articles/PMC9011180/.Dan","detector_url":"https://pith.science/pith-integrity-protocol#doi_compliance","external_url":null,"finding_type":"recoverable_identifier","evidence_hash":"b8d03dfd59b878f1b592284b1832dadd85cb34f6781460110274be6c9101e8dd","paper_version":1,"verdict_class":"incontrovertible","resolved_title":null,"detector_version":"1.0.0","detected_arxiv_id":null,"integrity_event_id":1029,"payload_sha256":"48678ac07ce6eaccc98321af3b7a35296aeb1fe98701da43f65e45aa1f840c05","signature_b64":"k5WUjEAwAhqL2GP6Jy30j0vBow/WBatq3OsYg9DFrsUWi8A1OY77ujWbbGuqj03YeUDcxtk8ziBqMBw7rThYCg==","signing_key_id":"pith-v1-2026-05"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-19T11:27:01Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"VCPph/tlAm+rKAMKo+Ac+rgEFtK/iFU7M1dYlj2OIvc6vB1OsFwKwIWTSbXVJ/flfWtwEXNukRaqiDLxcV5AAw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-02T10:20:35.914608Z"},"content_sha256":"9b8ad3dde762e54f4087e169628a6081c445719b4d17664fcb9de725e56fe71c","schema_version":"1.0","event_id":"sha256:9b8ad3dde762e54f4087e169628a6081c445719b4d17664fcb9de725e56fe71c"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/VQMPFD7EHMUMO6FVYUPLBU6POS/bundle.json","state_url":"https://pith.science/pith/VQMPFD7EHMUMO6FVYUPLBU6POS/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/VQMPFD7EHMUMO6FVYUPLBU6POS/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-06-02T10:20:35Z","links":{"resolver":"https://pith.science/pith/VQMPFD7EHMUMO6FVYUPLBU6POS","bundle":"https://pith.science/pith/VQMPFD7EHMUMO6FVYUPLBU6POS/bundle.json","state":"https://pith.science/pith/VQMPFD7EHMUMO6FVYUPLBU6POS/state.json","well_known_bundle":"https://pith.science/.well-known/pith/VQMPFD7EHMUMO6FVYUPLBU6POS/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:VQMPFD7EHMUMO6FVYUPLBU6POS","merge_version":"pith-open-graph-merge-v1","event_count":4,"valid_event_count":4,"invalid_event_count":0,"equivocation_count":1,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"133a9adea01e80b03ff81c4967815dba5865719a1437782d91a280f73ba236a3","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CV","submitted_at":"2026-05-08T15:55:30Z","title_canon_sha256":"53ac9af92f70e8a5751c4d6922585835fb90690dc00399e153d13ad726421550"},"schema_version":"1.0","source":{"id":"2605.07919","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.07919","created_at":"2026-05-25T02:02:16Z"},{"alias_kind":"arxiv_version","alias_value":"2605.07919v2","created_at":"2026-05-25T02:02:16Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.07919","created_at":"2026-05-25T02:02:16Z"},{"alias_kind":"pith_short_12","alias_value":"VQMPFD7EHMUM","created_at":"2026-05-25T02:02:16Z"},{"alias_kind":"pith_short_16","alias_value":"VQMPFD7EHMUMO6FV","created_at":"2026-05-25T02:02:16Z"},{"alias_kind":"pith_short_8","alias_value":"VQMPFD7E","created_at":"2026-05-25T02:02:16Z"}],"graph_snapshots":[{"event_id":"sha256:c3d4bac97121078b6e87557fb6ab519e093feb65cc4ce1c771e857c65c0ac20c","target":"graph","created_at":"2026-05-25T02:02:16Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"The independent radiologist scores MCS 83.3 at silent-failure rate 5.8%, leaving a 14.1-point composite headroom above the strongest audited model (Claude Opus 4.7 at 69.2)."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the four chosen perturbation types and the 300 clinician-authored cases sufficiently represent the range of broken visual evidence that occurs in real clinical practice."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"MedVIGIL introduces a clinician-supervised benchmark showing medical VLMs frequently give fluent answers on broken visual evidence, with top models 14 points below human radiologists on the composite score."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Medical vision-language models fail to refuse answers when visual evidence is broken, trailing radiologists by 14 points on a new composite score."}],"snapshot_sha256":"5abf56945b417f28f53c287fd507a3ccfd3d0c308bedf4408748d780b5d4b2a2"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":false,"detectors_run":[{"findings_count":0,"name":"claim_evidence","ran_at":"2026-05-20T10:02:13.555939Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"ai_meta_artifact","ran_at":"2026-05-20T04:46:22.489233Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_title_agreement","ran_at":"2026-05-19T15:31:18.310769Z","status":"completed","version":"1.0.0"},{"findings_count":2,"name":"doi_compliance","ran_at":"2026-05-19T11:24:20.741484Z","status":"completed","version":"1.0.0"}],"endpoint":"/pith/2605.07919/integrity.json","findings":[{"audited_at":"2026-05-19T11:24:20.741484Z","detected_arxiv_id":null,"detected_doi":"10.1109/TBME.2021.3117407.URLhttps://pmc.ncbi.nlm.nih.gov/articles/PMC9011180/.Dan","detector":"doi_compliance","finding_type":"recoverable_identifier","note":"DOI in the printed bibliography is fragmented by whitespace or line breaks. A longer candidate (10.1109/TBME.2021.3117407.URLhttps://pmc.ncbi.nlm.nih.gov/articles/PMC9011180/.Dan) was visible in the surrounding text but could not be confirmed against doi.org as printed.","ref_index":10,"severity":"advisory","verdict_class":"incontrovertible"},{"audited_at":"2026-05-19T11:24:20.741484Z","detected_arxiv_id":null,"detected_doi":"10.18653/v1/2023.emnlp-main.20.URLhttps://aclanthology.org/2023.emnlp-main.20/.Bo","detector":"doi_compliance","finding_type":"recoverable_identifier","note":"DOI in the printed bibliography is fragmented by whitespace or line breaks. A longer candidate (10.18653/v1/2023.emnlp-main.20.URLhttps://aclanthology.org/2023.emnlp-main.20/.Bo) was visible in the surrounding text but could not be confirmed against doi.org as printed.","ref_index":17,"severity":"advisory","verdict_class":"incontrovertible"}],"snapshot_sha256":"4abdae5894b4b9e13c60c8601c856f602b2de3eb7fdd7894a72fce302b49f1cc","summary":{"advisory":2,"by_detector":{"doi_compliance":{"advisory":2,"critical":0,"informational":0,"total":2}},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Medical vision--language models (VLMs) are usually evaluated on intact image--question pairs, but trustworthy clinical use requires a stronger property: a model must recognise when the evidential basis for an answer has failed. We study this through silent failures under perturbed evidence, where a vision-required medical question is paired with a false premise, wording perturbation, knowledge-only rewrite, or ROI-corrupted image, yet the model returns a fluent non-refusal answer. We introduce medvigil, a 300-case evaluation suite drawn from four public medical VQA sources, supervised end to e","authors_text":"Hanqi Jiang, Haozhen Gong, Hui Ren, Hyeokjae Kwon, Jinglei Lv, Junhao Chen, Lifeng Chen, Lin Zhao, Mingyu Kang, Quanzheng Li, Ruiyu Yan, Tianming Liu, Weihang You, Xiang Li, Yi Pan","cross_cats":[],"headline":"Medical vision-language models fail to refuse answers when visual evidence is broken, trailing radiologists by 14 points on a new composite score.","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CV","submitted_at":"2026-05-08T15:55:30Z","title":"MedVIGIL: Evaluating Trustworthy Medical VLMs Under Broken Visual Evidence"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.07919","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-11T03:40:45.192215Z","id":"40477877-05fe-4735-b79f-76bbcd099224","model_set":{"reader":"grok-4.3"},"one_line_summary":"MedVIGIL introduces a clinician-supervised benchmark showing medical VLMs frequently give fluent answers on broken visual evidence, with top models 14 points below human radiologists on the composite score.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Medical vision-language models fail to refuse answers when visual evidence is broken, trailing radiologists by 14 points on a new composite score.","strongest_claim":"The independent radiologist scores MCS 83.3 at silent-failure rate 5.8%, leaving a 14.1-point composite headroom above the strongest audited model (Claude Opus 4.7 at 69.2).","weakest_assumption":"That the four chosen perturbation types and the 300 clinician-authored cases sufficiently represent the range of broken visual evidence that occurs in real clinical practice."}},"verdict_id":"40477877-05fe-4735-b79f-76bbcd099224"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:d987f550ae76bdb920390ab8c740391d590ce01f56262d4c9ee595600804d93c","target":"record","created_at":"2026-05-25T02:02:16Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"133a9adea01e80b03ff81c4967815dba5865719a1437782d91a280f73ba236a3","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CV","submitted_at":"2026-05-08T15:55:30Z","title_canon_sha256":"53ac9af92f70e8a5751c4d6922585835fb90690dc00399e153d13ad726421550"},"schema_version":"1.0","source":{"id":"2605.07919","kind":"arxiv","version":2}},"canonical_sha256":"ac18f28fe43b28c778b5c51eb0d3cf74b15b8cf4701039e8417b656176d45bf0","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"ac18f28fe43b28c778b5c51eb0d3cf74b15b8cf4701039e8417b656176d45bf0","first_computed_at":"2026-05-25T02:02:16.508920Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-25T02:02:16.508920Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"vAlHsULuIJSPBomTOcKUcnXkUxCkUfgsQcWyKLCbEvTHKByTIlHvd0vHkhGqKT0K8tc9Usamc06CS1fzRGrVDA==","signature_status":"signed_v1","signed_at":"2026-05-25T02:02:16.509641Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.07919","source_kind":"arxiv","source_version":2}}},"equivocations":[{"signer_id":"pith.science","event_type":"integrity_finding","target":"integrity","event_ids":["sha256:9b8ad3dde762e54f4087e169628a6081c445719b4d17664fcb9de725e56fe71c","sha256:c7932fd0ad5b6493465116afc44e150f7ff7f89cde51d18a76123ed4553b00f1"]}],"invalid_events":[],"applied_event_ids":["sha256:d987f550ae76bdb920390ab8c740391d590ce01f56262d4c9ee595600804d93c","sha256:c3d4bac97121078b6e87557fb6ab519e093feb65cc4ce1c771e857c65c0ac20c"],"state_sha256":"30b14fb1cf434e7e2207fc8513d795fab3b2aeb68e22839216535fceba165f33"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"5Ka88ZDL3ABibc/bfnEAX8Tt+hp6JaL1k4b/qeRaQl5EMXylPPg4KcWPIzdOf2+czPo20w3e2Jke1kgU0J/QBw==","signed_message":"bundle_sha256_bytes","signed_at":"2026-06-02T10:20:35.917207Z","bundle_sha256":"8dc014e4e16a79bdd5c95ba799bc514a104c04ad01845050b99ec8e9bff0fc79"}}