{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:F2ULYW6KI4KEQT5EUDE6MEUVR4","short_pith_number":"pith:F2ULYW6K","schema_version":"1.0","canonical_sha256":"2ea8bc5bca4714484fa4a0c9e612958f36abfb9cb26d183ddab0dd705c7c88d6","source":{"kind":"arxiv","id":"2608.03744","version":1},"attestation_state":"computed","paper":{"title":"Agents Catching Agents: Shortcut Cascades and Benchmark Gaming in Clinical Multi-Agent Systems","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Aaditya Panchal, Agastya Munnangi, Aldo Marzullo, Anqi Peter Li, Armaan Grewal, Arshnoor Bhutani, Emmanuel Paul Kwesiga, Felipe Ocampo Osorio, Josephine Nanyonjo, Leo Anthony Celi, Maximin Lange, Milit S. Patel, Mohammad Shahin, Nikhil Jaiswal, Quang Bui, Sebasti\\'an Andr\\'es Cajas Ord\\'o\\~nez","submitted_at":"2026-08-04T14:37:16Z","abstract_excerpt":"Clinical decision support is moving toward committees of language-model agents deliberating on a shared workspace. We ask whether such committees can be gamed by shortcuts, cues a benchmark rewards but a clinician would ignore. Across seven cohorts on six public datasets spanning text (MedQA-USMLE, MedMCQA, MIMIC-CXR reports), imaging (NIH ChestX-ray14, MIMIC-CXR-JPG, CheXpert) and tabular ICU records (SUPPORT2), Gemini committees resist these cues in isolation (flip 5-16%), yet a socially plausible shortcut spreads: when two peers assert the same wrong answer, the holdout under test adopts it"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2608.03744","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-08-04T14:37:16Z","cross_cats_sorted":[],"title_canon_sha256":"ff8556f69964cca176701af4e7ea946f787bc01949e8222f49d9c1e69b286a79","abstract_canon_sha256":"bc94b604212a8e04d4dff856e74aafd13d0db1f8de809ee5b28796e11ec89471"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-08-05T01:37:00.762416Z","signature_b64":"PErESIvszZcprEQJppTNX/gH5bCLFIexvCDUjdEDvQp1bozGXy7mXR99GPijvqaHG4DEuftAzIbHoDQJrHINCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"2ea8bc5bca4714484fa4a0c9e612958f36abfb9cb26d183ddab0dd705c7c88d6","last_reissued_at":"2026-08-05T01:37:00.760975Z","signature_status":"signed_v1","first_computed_at":"2026-08-05T01:37:00.760975Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Agents Catching Agents: Shortcut Cascades and Benchmark Gaming in Clinical Multi-Agent Systems","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Aaditya Panchal, Agastya Munnangi, Aldo Marzullo, Anqi Peter Li, Armaan Grewal, Arshnoor Bhutani, Emmanuel Paul Kwesiga, Felipe Ocampo Osorio, Josephine Nanyonjo, Leo Anthony Celi, Maximin Lange, Milit S. Patel, Mohammad Shahin, Nikhil Jaiswal, Quang Bui, Sebasti\\'an Andr\\'es Cajas Ord\\'o\\~nez","submitted_at":"2026-08-04T14:37:16Z","abstract_excerpt":"Clinical decision support is moving toward committees of language-model agents deliberating on a shared workspace. We ask whether such committees can be gamed by shortcuts, cues a benchmark rewards but a clinician would ignore. Across seven cohorts on six public datasets spanning text (MedQA-USMLE, MedMCQA, MIMIC-CXR reports), imaging (NIH ChestX-ray14, MIMIC-CXR-JPG, CheXpert) and tabular ICU records (SUPPORT2), Gemini committees resist these cues in isolation (flip 5-16%), yet a socially plausible shortcut spreads: when two peers assert the same wrong answer, the holdout under test adopts it"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2608.03744","kind":"arxiv","version":1},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2608.03744/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"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"},"aliases":[{"alias_kind":"arxiv","alias_value":"2608.03744","created_at":"2026-08-05T01:37:00.761409+00:00"},{"alias_kind":"arxiv_version","alias_value":"2608.03744v1","created_at":"2026-08-05T01:37:00.761409+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2608.03744","created_at":"2026-08-05T01:37:00.761409+00:00"},{"alias_kind":"pith_short_12","alias_value":"F2ULYW6KI4KE","created_at":"2026-08-05T01:37:00.761409+00:00"},{"alias_kind":"pith_short_16","alias_value":"F2ULYW6KI4KEQT5E","created_at":"2026-08-05T01:37:00.761409+00:00"},{"alias_kind":"pith_short_8","alias_value":"F2ULYW6K","created_at":"2026-08-05T01:37:00.761409+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/F2ULYW6KI4KEQT5EUDE6MEUVR4","json":"https://pith.science/pith/F2ULYW6KI4KEQT5EUDE6MEUVR4.json","graph_json":"https://pith.science/api/pith-number/F2ULYW6KI4KEQT5EUDE6MEUVR4/graph.json","events_json":"https://pith.science/api/pith-number/F2ULYW6KI4KEQT5EUDE6MEUVR4/events.json","paper":"https://pith.science/paper/F2ULYW6K"},"agent_actions":{"view_html":"https://pith.science/pith/F2ULYW6KI4KEQT5EUDE6MEUVR4","download_json":"https://pith.science/pith/F2ULYW6KI4KEQT5EUDE6MEUVR4.json","view_paper":"https://pith.science/paper/F2ULYW6K","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2608.03744&json=true","fetch_graph":"https://pith.science/api/pith-number/F2ULYW6KI4KEQT5EUDE6MEUVR4/graph.json","fetch_events":"https://pith.science/api/pith-number/F2ULYW6KI4KEQT5EUDE6MEUVR4/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/F2ULYW6KI4KEQT5EUDE6MEUVR4/action/timestamp_anchor","attest_storage":"https://pith.science/pith/F2ULYW6KI4KEQT5EUDE6MEUVR4/action/storage_attestation","attest_author":"https://pith.science/pith/F2ULYW6KI4KEQT5EUDE6MEUVR4/action/author_attestation","sign_citation":"https://pith.science/pith/F2ULYW6KI4KEQT5EUDE6MEUVR4/action/citation_signature","submit_replication":"https://pith.science/pith/F2ULYW6KI4KEQT5EUDE6MEUVR4/action/replication_record"}},"created_at":"2026-08-05T01:37:00.761409+00:00","updated_at":"2026-08-05T01:37:00.761409+00:00"}