{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:F6XERQ72HKF42FGFUJLZRJWDQ4","short_pith_number":"pith:F6XERQ72","schema_version":"1.0","canonical_sha256":"2fae48c3fa3a8bcd14c5a25798a6c3872e7bbf457c9d3b9a95136d66135fe3ae","source":{"kind":"arxiv","id":"2605.31381","version":1},"attestation_state":"computed","paper":{"title":"LLM Judges Inconsistently Disagree Across Safety Criteria and Harm Categories","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Akriti Vij, Isar Nejadgholi, Krishnapriya Vishnubhotla, Soumya Vajjala","submitted_at":"2026-05-29T14:50:08Z","abstract_excerpt":"We evaluate the consistency of automated judges in conducting a multi-dimensional safety evaluation in a reference-free setup. Our results indicate that Large Language Models are unreliable judges in identifying safety issues related to machine-generated advice in regulated domains such as finance, although they are more reliable at identifying more overt forms of unsafe/harmful content such as violence. The degree of inconsistency in a model's judgments can vary significantly by the chosen safety criteria and can be impacted by the language of the content and its linguistic style as well. Fin"},"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":"2605.31381","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-29T14:50:08Z","cross_cats_sorted":[],"title_canon_sha256":"31121cc6c946e328f51c177179993d070977c03e0f06b4e866b326cd6c7f7bb2","abstract_canon_sha256":"5b7af69b25227fe8e60ac942c298b064bc5edf0823a8916909ac4fe79e0fb3e3"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-01T02:04:02.140670Z","signature_b64":"9KeZJdlSuBA7O9UxfdK2rs5hHND+8jXDZLLiCE1R7654RfuNF21wGfV3kgqn3WXqclRZWehekTPoHHq2t7GGCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"2fae48c3fa3a8bcd14c5a25798a6c3872e7bbf457c9d3b9a95136d66135fe3ae","last_reissued_at":"2026-06-01T02:04:02.139551Z","signature_status":"signed_v1","first_computed_at":"2026-06-01T02:04:02.139551Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"LLM Judges Inconsistently Disagree Across Safety Criteria and Harm Categories","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Akriti Vij, Isar Nejadgholi, Krishnapriya Vishnubhotla, Soumya Vajjala","submitted_at":"2026-05-29T14:50:08Z","abstract_excerpt":"We evaluate the consistency of automated judges in conducting a multi-dimensional safety evaluation in a reference-free setup. Our results indicate that Large Language Models are unreliable judges in identifying safety issues related to machine-generated advice in regulated domains such as finance, although they are more reliable at identifying more overt forms of unsafe/harmful content such as violence. The degree of inconsistency in a model's judgments can vary significantly by the chosen safety criteria and can be impacted by the language of the content and its linguistic style as well. Fin"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.31381","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/2605.31381/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":"2605.31381","created_at":"2026-06-01T02:04:02.139761+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.31381v1","created_at":"2026-06-01T02:04:02.139761+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.31381","created_at":"2026-06-01T02:04:02.139761+00:00"},{"alias_kind":"pith_short_12","alias_value":"F6XERQ72HKF4","created_at":"2026-06-01T02:04:02.139761+00:00"},{"alias_kind":"pith_short_16","alias_value":"F6XERQ72HKF42FGF","created_at":"2026-06-01T02:04:02.139761+00:00"},{"alias_kind":"pith_short_8","alias_value":"F6XERQ72","created_at":"2026-06-01T02:04:02.139761+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/F6XERQ72HKF42FGFUJLZRJWDQ4","json":"https://pith.science/pith/F6XERQ72HKF42FGFUJLZRJWDQ4.json","graph_json":"https://pith.science/api/pith-number/F6XERQ72HKF42FGFUJLZRJWDQ4/graph.json","events_json":"https://pith.science/api/pith-number/F6XERQ72HKF42FGFUJLZRJWDQ4/events.json","paper":"https://pith.science/paper/F6XERQ72"},"agent_actions":{"view_html":"https://pith.science/pith/F6XERQ72HKF42FGFUJLZRJWDQ4","download_json":"https://pith.science/pith/F6XERQ72HKF42FGFUJLZRJWDQ4.json","view_paper":"https://pith.science/paper/F6XERQ72","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.31381&json=true","fetch_graph":"https://pith.science/api/pith-number/F6XERQ72HKF42FGFUJLZRJWDQ4/graph.json","fetch_events":"https://pith.science/api/pith-number/F6XERQ72HKF42FGFUJLZRJWDQ4/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/F6XERQ72HKF42FGFUJLZRJWDQ4/action/timestamp_anchor","attest_storage":"https://pith.science/pith/F6XERQ72HKF42FGFUJLZRJWDQ4/action/storage_attestation","attest_author":"https://pith.science/pith/F6XERQ72HKF42FGFUJLZRJWDQ4/action/author_attestation","sign_citation":"https://pith.science/pith/F6XERQ72HKF42FGFUJLZRJWDQ4/action/citation_signature","submit_replication":"https://pith.science/pith/F6XERQ72HKF42FGFUJLZRJWDQ4/action/replication_record"}},"created_at":"2026-06-01T02:04:02.139761+00:00","updated_at":"2026-06-01T02:04:02.139761+00:00"}