{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:FPBH3VH44UGGU3QQE7DQCLUTNB","short_pith_number":"pith:FPBH3VH4","schema_version":"1.0","canonical_sha256":"2bc27dd4fce50c6a6e1027c7012e936863751ef3c806acd6bee2f21a1066deba","source":{"kind":"arxiv","id":"2605.20759","version":1},"attestation_state":"computed","paper":{"title":"Rethinking Fraud Safety Evaluation: Multi-Round Attacks Reveal Safety-Utility Tradeoffs in Graph-Context LLM Defenders","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CR","authors_text":"Laura Jiang, Nasim Ferdosian, Qian Li, Reza Ryan","submitted_at":"2026-05-20T05:59:54Z","abstract_excerpt":"Single-turn safety evaluation is a poor proxy for real fraud defense, where attackers escalate across multiple rounds. This paper evaluates fraud defenders under replay and adaptive multi-round attacks and measures when a defender refuses, not just whether it eventually refuses. On a frozen multi-round suite built from Fraud-R1, graph-context defenders improve early safe refusal relative to text-only baselines under both replay and adaptive fraud pressure, but they also produce substantially more benign over-refusal. Direct probing of the trained graph encoder, together with paired shuffle-ris"},"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":"2605.20759","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CR","submitted_at":"2026-05-20T05:59:54Z","cross_cats_sorted":[],"title_canon_sha256":"e41793898d9c5dda4dd6d483cc6e16dc4d9a1991e5f0fee888c1915c3136b816","abstract_canon_sha256":"e5ded62b60d4b46e4f0b532081ac38de77f9be50aaccd3b76ba5512adf186a72"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-21T01:04:52.841556Z","signature_b64":"ZnEGwb32anOy78V5ik55n7+h0lS4wyCS+feNTWNWC/LxOdeuCp+RUmiQe4eIoLwmtecG2pJ9I4SxS4yovXslBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"2bc27dd4fce50c6a6e1027c7012e936863751ef3c806acd6bee2f21a1066deba","last_reissued_at":"2026-05-21T01:04:52.840768Z","signature_status":"signed_v1","first_computed_at":"2026-05-21T01:04:52.840768Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Rethinking Fraud Safety Evaluation: Multi-Round Attacks Reveal Safety-Utility Tradeoffs in Graph-Context LLM Defenders","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CR","authors_text":"Laura Jiang, Nasim Ferdosian, Qian Li, Reza Ryan","submitted_at":"2026-05-20T05:59:54Z","abstract_excerpt":"Single-turn safety evaluation is a poor proxy for real fraud defense, where attackers escalate across multiple rounds. This paper evaluates fraud defenders under replay and adaptive multi-round attacks and measures when a defender refuses, not just whether it eventually refuses. On a frozen multi-round suite built from Fraud-R1, graph-context defenders improve early safe refusal relative to text-only baselines under both replay and adaptive fraud pressure, but they also produce substantially more benign over-refusal. Direct probing of the trained graph encoder, together with paired shuffle-ris"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.20759","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/2605.20759/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":"2605.20759","created_at":"2026-05-21T01:04:52.840899+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.20759v1","created_at":"2026-05-21T01:04:52.840899+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.20759","created_at":"2026-05-21T01:04:52.840899+00:00"},{"alias_kind":"pith_short_12","alias_value":"FPBH3VH44UGG","created_at":"2026-05-21T01:04:52.840899+00:00"},{"alias_kind":"pith_short_16","alias_value":"FPBH3VH44UGGU3QQ","created_at":"2026-05-21T01:04:52.840899+00:00"},{"alias_kind":"pith_short_8","alias_value":"FPBH3VH4","created_at":"2026-05-21T01:04:52.840899+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/FPBH3VH44UGGU3QQE7DQCLUTNB","json":"https://pith.science/pith/FPBH3VH44UGGU3QQE7DQCLUTNB.json","graph_json":"https://pith.science/api/pith-number/FPBH3VH44UGGU3QQE7DQCLUTNB/graph.json","events_json":"https://pith.science/api/pith-number/FPBH3VH44UGGU3QQE7DQCLUTNB/events.json","paper":"https://pith.science/paper/FPBH3VH4"},"agent_actions":{"view_html":"https://pith.science/pith/FPBH3VH44UGGU3QQE7DQCLUTNB","download_json":"https://pith.science/pith/FPBH3VH44UGGU3QQE7DQCLUTNB.json","view_paper":"https://pith.science/paper/FPBH3VH4","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.20759&json=true","fetch_graph":"https://pith.science/api/pith-number/FPBH3VH44UGGU3QQE7DQCLUTNB/graph.json","fetch_events":"https://pith.science/api/pith-number/FPBH3VH44UGGU3QQE7DQCLUTNB/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/FPBH3VH44UGGU3QQE7DQCLUTNB/action/timestamp_anchor","attest_storage":"https://pith.science/pith/FPBH3VH44UGGU3QQE7DQCLUTNB/action/storage_attestation","attest_author":"https://pith.science/pith/FPBH3VH44UGGU3QQE7DQCLUTNB/action/author_attestation","sign_citation":"https://pith.science/pith/FPBH3VH44UGGU3QQE7DQCLUTNB/action/citation_signature","submit_replication":"https://pith.science/pith/FPBH3VH44UGGU3QQE7DQCLUTNB/action/replication_record"}},"created_at":"2026-05-21T01:04:52.840899+00:00","updated_at":"2026-05-21T01:04:52.840899+00:00"}