{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:DSFIX7DOOETYKDUM2GTK2F5ZHT","short_pith_number":"pith:DSFIX7DO","schema_version":"1.0","canonical_sha256":"1c8a8bfc6e7127850e8cd1a6ad17b93ceebc75a46a18174972ac2ed5f305d9c6","source":{"kind":"arxiv","id":"2503.04474","version":1},"attestation_state":"computed","paper":{"title":"Know Thy Judge: On the Robustness Meta-Evaluation of LLM Safety Judges","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CR"],"primary_cat":"cs.LG","authors_text":"Eliott Zemour, Eric Lin, Francisco Eiras, Vaikkunth Mugunthan","submitted_at":"2025-03-06T14:24:12Z","abstract_excerpt":"Large Language Model (LLM) based judges form the underpinnings of key safety evaluation processes such as offline benchmarking, automated red-teaming, and online guardrailing. This widespread requirement raises the crucial question: can we trust the evaluations of these evaluators? In this paper, we highlight two critical challenges that are typically overlooked: (i) evaluations in the wild where factors like prompt sensitivity and distribution shifts can affect performance and (ii) adversarial attacks that target the judge. We highlight the importance of these through a study of commonly used"},"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":"2503.04474","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-03-06T14:24:12Z","cross_cats_sorted":["cs.CR"],"title_canon_sha256":"2b564964d75f4d7db32ac3b109fde28cf91be2a9868976d80781745addca0448","abstract_canon_sha256":"b7cd727e9a5c4503259120b70aabd5aa8853862d1eda76772ac1c8b058109304"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:25:39.614503Z","signature_b64":"jtSN7WumeN34KNAGDA48CV6dInyJq3GpcKJ/eM7xQ8ZAYj493eGVIPeZsO0cqWBf5j35T1IMkUtRr2J2+bC5DA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1c8a8bfc6e7127850e8cd1a6ad17b93ceebc75a46a18174972ac2ed5f305d9c6","last_reissued_at":"2026-07-05T10:25:39.614009Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:25:39.614009Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Know Thy Judge: On the Robustness Meta-Evaluation of LLM Safety Judges","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CR"],"primary_cat":"cs.LG","authors_text":"Eliott Zemour, Eric Lin, Francisco Eiras, Vaikkunth Mugunthan","submitted_at":"2025-03-06T14:24:12Z","abstract_excerpt":"Large Language Model (LLM) based judges form the underpinnings of key safety evaluation processes such as offline benchmarking, automated red-teaming, and online guardrailing. This widespread requirement raises the crucial question: can we trust the evaluations of these evaluators? In this paper, we highlight two critical challenges that are typically overlooked: (i) evaluations in the wild where factors like prompt sensitivity and distribution shifts can affect performance and (ii) adversarial attacks that target the judge. We highlight the importance of these through a study of commonly used"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.04474","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/2503.04474/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":"2503.04474","created_at":"2026-07-05T10:25:39.614068+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.04474v1","created_at":"2026-07-05T10:25:39.614068+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.04474","created_at":"2026-07-05T10:25:39.614068+00:00"},{"alias_kind":"pith_short_12","alias_value":"DSFIX7DOOETY","created_at":"2026-07-05T10:25:39.614068+00:00"},{"alias_kind":"pith_short_16","alias_value":"DSFIX7DOOETYKDUM","created_at":"2026-07-05T10:25:39.614068+00:00"},{"alias_kind":"pith_short_8","alias_value":"DSFIX7DO","created_at":"2026-07-05T10:25:39.614068+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.25487","citing_title":"How Reliable Is Your Jailbreak Judge? Calibration and Adversarial Robustness of Automated ASR Scoring","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2606.25487","citing_title":"How Reliable Is Your Jailbreak Judge? Calibration and Adversarial Robustness of Automated ASR Scoring","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2606.09165","citing_title":"Reliable to Expressive: A Curriculum for Rubric-Following Safety Judges","ref_index":2,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/DSFIX7DOOETYKDUM2GTK2F5ZHT","json":"https://pith.science/pith/DSFIX7DOOETYKDUM2GTK2F5ZHT.json","graph_json":"https://pith.science/api/pith-number/DSFIX7DOOETYKDUM2GTK2F5ZHT/graph.json","events_json":"https://pith.science/api/pith-number/DSFIX7DOOETYKDUM2GTK2F5ZHT/events.json","paper":"https://pith.science/paper/DSFIX7DO"},"agent_actions":{"view_html":"https://pith.science/pith/DSFIX7DOOETYKDUM2GTK2F5ZHT","download_json":"https://pith.science/pith/DSFIX7DOOETYKDUM2GTK2F5ZHT.json","view_paper":"https://pith.science/paper/DSFIX7DO","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.04474&json=true","fetch_graph":"https://pith.science/api/pith-number/DSFIX7DOOETYKDUM2GTK2F5ZHT/graph.json","fetch_events":"https://pith.science/api/pith-number/DSFIX7DOOETYKDUM2GTK2F5ZHT/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/DSFIX7DOOETYKDUM2GTK2F5ZHT/action/timestamp_anchor","attest_storage":"https://pith.science/pith/DSFIX7DOOETYKDUM2GTK2F5ZHT/action/storage_attestation","attest_author":"https://pith.science/pith/DSFIX7DOOETYKDUM2GTK2F5ZHT/action/author_attestation","sign_citation":"https://pith.science/pith/DSFIX7DOOETYKDUM2GTK2F5ZHT/action/citation_signature","submit_replication":"https://pith.science/pith/DSFIX7DOOETYKDUM2GTK2F5ZHT/action/replication_record"}},"created_at":"2026-07-05T10:25:39.614068+00:00","updated_at":"2026-07-05T10:25:39.614068+00:00"}