{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:W424NIQGYCGE2PDPLXMSVTCSDS","short_pith_number":"pith:W424NIQG","schema_version":"1.0","canonical_sha256":"b735c6a206c08c4d3c6f5dd92acc521c988bc6f31eb6f520dc3bdcb17184b057","source":{"kind":"arxiv","id":"2410.21965","version":1},"attestation_state":"computed","paper":{"title":"SG-Bench: Evaluating LLM Safety Generalization Across Diverse Tasks and Prompt Types","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Shikun Zhang, Wei Ye, Yutao Mou","submitted_at":"2024-10-29T11:47:01Z","abstract_excerpt":"Ensuring the safety of large language model (LLM) applications is essential for developing trustworthy artificial intelligence. Current LLM safety benchmarks have two limitations. First, they focus solely on either discriminative or generative evaluation paradigms while ignoring their interconnection. Second, they rely on standardized inputs, overlooking the effects of widespread prompting techniques, such as system prompts, few-shot demonstrations, and chain-of-thought prompting. To overcome these issues, we developed SG-Bench, a novel benchmark to assess the generalization of LLM safety acro"},"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":"2410.21965","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-10-29T11:47:01Z","cross_cats_sorted":[],"title_canon_sha256":"0d3c72e13435c0d19acb2202c12fe6f7c5187968e1ed1ef72e090b9a013ffea2","abstract_canon_sha256":"0a5f34124c09b673686bd2a4e173ecc7b2a4ffd18285608d60b50890bb643488"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:28:02.263608Z","signature_b64":"cLxNBlQexEfDlF+h0NQMbuOqcxvd6yvkRptq4gXhmzN4/BIhGalHSVbCP2Ml2xjUhiqBTbdPYiEgPUuarYYlCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b735c6a206c08c4d3c6f5dd92acc521c988bc6f31eb6f520dc3bdcb17184b057","last_reissued_at":"2026-07-05T09:28:02.263019Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:28:02.263019Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"SG-Bench: Evaluating LLM Safety Generalization Across Diverse Tasks and Prompt Types","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Shikun Zhang, Wei Ye, Yutao Mou","submitted_at":"2024-10-29T11:47:01Z","abstract_excerpt":"Ensuring the safety of large language model (LLM) applications is essential for developing trustworthy artificial intelligence. Current LLM safety benchmarks have two limitations. First, they focus solely on either discriminative or generative evaluation paradigms while ignoring their interconnection. Second, they rely on standardized inputs, overlooking the effects of widespread prompting techniques, such as system prompts, few-shot demonstrations, and chain-of-thought prompting. To overcome these issues, we developed SG-Bench, a novel benchmark to assess the generalization of LLM safety acro"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.21965","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/2410.21965/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":"2410.21965","created_at":"2026-07-05T09:28:02.263106+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.21965v1","created_at":"2026-07-05T09:28:02.263106+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.21965","created_at":"2026-07-05T09:28:02.263106+00:00"},{"alias_kind":"pith_short_12","alias_value":"W424NIQGYCGE","created_at":"2026-07-05T09:28:02.263106+00:00"},{"alias_kind":"pith_short_16","alias_value":"W424NIQGYCGE2PDP","created_at":"2026-07-05T09:28:02.263106+00:00"},{"alias_kind":"pith_short_8","alias_value":"W424NIQG","created_at":"2026-07-05T09:28:02.263106+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.30049","citing_title":"Robust and Generalizable Safety Steering for Text-to-Image Diffusion Transformers","ref_index":34,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20351","citing_title":"Refusal Evaluation in Coding LLMs and Code Agents: A Systematic Review of Thirteen Malicious-Code Prompt Corpora (2023-2025)","ref_index":33,"is_internal_anchor":false},{"citing_arxiv_id":"2604.14548","citing_title":"VoxSafeBench: Not Just What Is Said, but Who, How, and Where","ref_index":16,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/W424NIQGYCGE2PDPLXMSVTCSDS","json":"https://pith.science/pith/W424NIQGYCGE2PDPLXMSVTCSDS.json","graph_json":"https://pith.science/api/pith-number/W424NIQGYCGE2PDPLXMSVTCSDS/graph.json","events_json":"https://pith.science/api/pith-number/W424NIQGYCGE2PDPLXMSVTCSDS/events.json","paper":"https://pith.science/paper/W424NIQG"},"agent_actions":{"view_html":"https://pith.science/pith/W424NIQGYCGE2PDPLXMSVTCSDS","download_json":"https://pith.science/pith/W424NIQGYCGE2PDPLXMSVTCSDS.json","view_paper":"https://pith.science/paper/W424NIQG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.21965&json=true","fetch_graph":"https://pith.science/api/pith-number/W424NIQGYCGE2PDPLXMSVTCSDS/graph.json","fetch_events":"https://pith.science/api/pith-number/W424NIQGYCGE2PDPLXMSVTCSDS/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/W424NIQGYCGE2PDPLXMSVTCSDS/action/timestamp_anchor","attest_storage":"https://pith.science/pith/W424NIQGYCGE2PDPLXMSVTCSDS/action/storage_attestation","attest_author":"https://pith.science/pith/W424NIQGYCGE2PDPLXMSVTCSDS/action/author_attestation","sign_citation":"https://pith.science/pith/W424NIQGYCGE2PDPLXMSVTCSDS/action/citation_signature","submit_replication":"https://pith.science/pith/W424NIQGYCGE2PDPLXMSVTCSDS/action/replication_record"}},"created_at":"2026-07-05T09:28:02.263106+00:00","updated_at":"2026-07-05T09:28:02.263106+00:00"}