{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:4CRWGGVF476MI52JK6MUBSABHF","short_pith_number":"pith:4CRWGGVF","schema_version":"1.0","canonical_sha256":"e0a3631aa5e7fcc47749579940c80139629c57d09f2a62828c83bdd507663b79","source":{"kind":"arxiv","id":"2606.07874","version":1},"attestation_state":"computed","paper":{"title":"Safety is Contextual, LLM-Judges Are Not: Navigating the Rigid Priors of Evaluators","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Anissa Alloula, Ava Batchkala, Federico Licini, Seraphina Goldfarb-Tarrant","submitted_at":"2026-06-05T22:11:26Z","abstract_excerpt":"LLMs-as-judges are the only way to evaluate safety at scale. Despite their importance, LLM-judges themselves are rarely evaluated beyond human agreement in simple, static benchmarks. We therefore investigate two under-explored but crucial properties of LLMs-as-judges: their susceptibility to relying on in context-information, and their steerability to differing safety definitions, which may not align with their internal safety priors. We evaluate the safety judging abilities of many generalist LLMs and safety-specific judges, and investigate the impact of task demonstrations, novel in-context "},"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":"2606.07874","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-06-05T22:11:26Z","cross_cats_sorted":[],"title_canon_sha256":"0d14ae625adf38eb42fb57dc56f88cc89ca99e0ba726625b246a5ce592a740c5","abstract_canon_sha256":"b49e792501d1e8a2c9dab8842c915f9d87a16640a250aa46af45fd39ba7702f3"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-09T01:04:54.287859Z","signature_b64":"X+jRXDS6takH0vj34l67d8sxES9LdtQjrqGN65kPYz5uKJ5W3gpUatdZ+AYus6VzA86yaafaknPFoqa5Y0HAAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e0a3631aa5e7fcc47749579940c80139629c57d09f2a62828c83bdd507663b79","last_reissued_at":"2026-06-09T01:04:54.287492Z","signature_status":"signed_v1","first_computed_at":"2026-06-09T01:04:54.287492Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Safety is Contextual, LLM-Judges Are Not: Navigating the Rigid Priors of Evaluators","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Anissa Alloula, Ava Batchkala, Federico Licini, Seraphina Goldfarb-Tarrant","submitted_at":"2026-06-05T22:11:26Z","abstract_excerpt":"LLMs-as-judges are the only way to evaluate safety at scale. Despite their importance, LLM-judges themselves are rarely evaluated beyond human agreement in simple, static benchmarks. We therefore investigate two under-explored but crucial properties of LLMs-as-judges: their susceptibility to relying on in context-information, and their steerability to differing safety definitions, which may not align with their internal safety priors. We evaluate the safety judging abilities of many generalist LLMs and safety-specific judges, and investigate the impact of task demonstrations, novel in-context "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.07874","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/2606.07874/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":"2606.07874","created_at":"2026-06-09T01:04:54.287554+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.07874v1","created_at":"2026-06-09T01:04:54.287554+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.07874","created_at":"2026-06-09T01:04:54.287554+00:00"},{"alias_kind":"pith_short_12","alias_value":"4CRWGGVF476M","created_at":"2026-06-09T01:04:54.287554+00:00"},{"alias_kind":"pith_short_16","alias_value":"4CRWGGVF476MI52J","created_at":"2026-06-09T01:04:54.287554+00:00"},{"alias_kind":"pith_short_8","alias_value":"4CRWGGVF","created_at":"2026-06-09T01:04:54.287554+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/4CRWGGVF476MI52JK6MUBSABHF","json":"https://pith.science/pith/4CRWGGVF476MI52JK6MUBSABHF.json","graph_json":"https://pith.science/api/pith-number/4CRWGGVF476MI52JK6MUBSABHF/graph.json","events_json":"https://pith.science/api/pith-number/4CRWGGVF476MI52JK6MUBSABHF/events.json","paper":"https://pith.science/paper/4CRWGGVF"},"agent_actions":{"view_html":"https://pith.science/pith/4CRWGGVF476MI52JK6MUBSABHF","download_json":"https://pith.science/pith/4CRWGGVF476MI52JK6MUBSABHF.json","view_paper":"https://pith.science/paper/4CRWGGVF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.07874&json=true","fetch_graph":"https://pith.science/api/pith-number/4CRWGGVF476MI52JK6MUBSABHF/graph.json","fetch_events":"https://pith.science/api/pith-number/4CRWGGVF476MI52JK6MUBSABHF/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/4CRWGGVF476MI52JK6MUBSABHF/action/timestamp_anchor","attest_storage":"https://pith.science/pith/4CRWGGVF476MI52JK6MUBSABHF/action/storage_attestation","attest_author":"https://pith.science/pith/4CRWGGVF476MI52JK6MUBSABHF/action/author_attestation","sign_citation":"https://pith.science/pith/4CRWGGVF476MI52JK6MUBSABHF/action/citation_signature","submit_replication":"https://pith.science/pith/4CRWGGVF476MI52JK6MUBSABHF/action/replication_record"}},"created_at":"2026-06-09T01:04:54.287554+00:00","updated_at":"2026-06-09T01:04:54.287554+00:00"}