{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:DJBI4SCTW4RQNRTYKREECMCZOZ","short_pith_number":"pith:DJBI4SCT","schema_version":"1.0","canonical_sha256":"1a428e4853b72306c6785448413059767edeac296d6a50e761fa76b53e969c90","source":{"kind":"arxiv","id":"2605.28030","version":1},"attestation_state":"computed","paper":{"title":"SPARD: Defending Harmful Fine-Tuning Attack via Safety Projection with Relevance-Diversity Data Selection","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CR"],"primary_cat":"cs.LG","authors_text":"Chengxiang Zhuo, James T. Kwok, Shengda Luo, Shuhao Chen, Weisen Jiang, Yeqi Gong, Yu Zhang, Zang Li","submitted_at":"2026-05-27T06:36:22Z","abstract_excerpt":"Fine-tuning large language models often undermines their safety alignment, a problem further amplified by harmful fine-tuning attacks in which adversarial data removes safeguards and induces unsafe behaviors. We propose SPARD, a defense framework that integrates Safety-Projected Alternating optimization with Relevance-Diversity aware data selection. SPARD employs SPAG, which optimizes alternatively between utility updates and explicit safety projections with a set of safe data to enforce safety constraints. To curate safe data, we introduce a Relevance-Diversity Determinantal Point Process to "},"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":"2605.28030","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-27T06:36:22Z","cross_cats_sorted":["cs.AI","cs.CR"],"title_canon_sha256":"9c206e5fd9de4446ad7ff16ffec85fc31513e7c826326b15835e218b82b0e34a","abstract_canon_sha256":"9d1f0e7d16fb70f33a8145e6c5a1e164d0983e0650cc28ee0baa03cd8e2e6230"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-28T01:04:56.482625Z","signature_b64":"b+DmOIxY9XhRT1krhSpOKNWqpCKzU3G0Ebg5ZpsrSOOVZkEmeWSI62fyv+yUKIn2WL0SBzzXkFkxWb8Qr2VRAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1a428e4853b72306c6785448413059767edeac296d6a50e761fa76b53e969c90","last_reissued_at":"2026-05-28T01:04:56.482226Z","signature_status":"signed_v1","first_computed_at":"2026-05-28T01:04:56.482226Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"SPARD: Defending Harmful Fine-Tuning Attack via Safety Projection with Relevance-Diversity Data Selection","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CR"],"primary_cat":"cs.LG","authors_text":"Chengxiang Zhuo, James T. Kwok, Shengda Luo, Shuhao Chen, Weisen Jiang, Yeqi Gong, Yu Zhang, Zang Li","submitted_at":"2026-05-27T06:36:22Z","abstract_excerpt":"Fine-tuning large language models often undermines their safety alignment, a problem further amplified by harmful fine-tuning attacks in which adversarial data removes safeguards and induces unsafe behaviors. We propose SPARD, a defense framework that integrates Safety-Projected Alternating optimization with Relevance-Diversity aware data selection. SPARD employs SPAG, which optimizes alternatively between utility updates and explicit safety projections with a set of safe data to enforce safety constraints. To curate safe data, we introduce a Relevance-Diversity Determinantal Point Process to "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.28030","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/2605.28030/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":"2605.28030","created_at":"2026-05-28T01:04:56.482278+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.28030v1","created_at":"2026-05-28T01:04:56.482278+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.28030","created_at":"2026-05-28T01:04:56.482278+00:00"},{"alias_kind":"pith_short_12","alias_value":"DJBI4SCTW4RQ","created_at":"2026-05-28T01:04:56.482278+00:00"},{"alias_kind":"pith_short_16","alias_value":"DJBI4SCTW4RQNRTY","created_at":"2026-05-28T01:04:56.482278+00:00"},{"alias_kind":"pith_short_8","alias_value":"DJBI4SCT","created_at":"2026-05-28T01:04:56.482278+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/DJBI4SCTW4RQNRTYKREECMCZOZ","json":"https://pith.science/pith/DJBI4SCTW4RQNRTYKREECMCZOZ.json","graph_json":"https://pith.science/api/pith-number/DJBI4SCTW4RQNRTYKREECMCZOZ/graph.json","events_json":"https://pith.science/api/pith-number/DJBI4SCTW4RQNRTYKREECMCZOZ/events.json","paper":"https://pith.science/paper/DJBI4SCT"},"agent_actions":{"view_html":"https://pith.science/pith/DJBI4SCTW4RQNRTYKREECMCZOZ","download_json":"https://pith.science/pith/DJBI4SCTW4RQNRTYKREECMCZOZ.json","view_paper":"https://pith.science/paper/DJBI4SCT","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.28030&json=true","fetch_graph":"https://pith.science/api/pith-number/DJBI4SCTW4RQNRTYKREECMCZOZ/graph.json","fetch_events":"https://pith.science/api/pith-number/DJBI4SCTW4RQNRTYKREECMCZOZ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/DJBI4SCTW4RQNRTYKREECMCZOZ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/DJBI4SCTW4RQNRTYKREECMCZOZ/action/storage_attestation","attest_author":"https://pith.science/pith/DJBI4SCTW4RQNRTYKREECMCZOZ/action/author_attestation","sign_citation":"https://pith.science/pith/DJBI4SCTW4RQNRTYKREECMCZOZ/action/citation_signature","submit_replication":"https://pith.science/pith/DJBI4SCTW4RQNRTYKREECMCZOZ/action/replication_record"}},"created_at":"2026-05-28T01:04:56.482278+00:00","updated_at":"2026-05-28T01:04:56.482278+00:00"}