{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2024:BXY6Z5YIIIUPOIUYI24OEYINHR","short_pith_number":"pith:BXY6Z5YI","canonical_record":{"source":{"id":"2410.07471","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-10-09T22:24:22Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"2fcabf01881c1e39ab9b9a722376a470b0732eefd592682684fecc72fba29221","abstract_canon_sha256":"b51dceff7c7b916fa158865bcabf3830a8aa7a2438c1dfac665d497fb43f86d0"},"schema_version":"1.0"},"canonical_sha256":"0df1ecf7084228f7229846b8e2610d3c47610f9b152ee5756b5145ef0ef0fadd","source":{"kind":"arxiv","id":"2410.07471","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2410.07471","created_at":"2026-07-05T09:18:54Z"},{"alias_kind":"arxiv_version","alias_value":"2410.07471v2","created_at":"2026-07-05T09:18:54Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.07471","created_at":"2026-07-05T09:18:54Z"},{"alias_kind":"pith_short_12","alias_value":"BXY6Z5YIIIUP","created_at":"2026-07-05T09:18:54Z"},{"alias_kind":"pith_short_16","alias_value":"BXY6Z5YIIIUPOIUY","created_at":"2026-07-05T09:18:54Z"},{"alias_kind":"pith_short_8","alias_value":"BXY6Z5YI","created_at":"2026-07-05T09:18:54Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2024:BXY6Z5YIIIUPOIUYI24OEYINHR","target":"record","payload":{"canonical_record":{"source":{"id":"2410.07471","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-10-09T22:24:22Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"2fcabf01881c1e39ab9b9a722376a470b0732eefd592682684fecc72fba29221","abstract_canon_sha256":"b51dceff7c7b916fa158865bcabf3830a8aa7a2438c1dfac665d497fb43f86d0"},"schema_version":"1.0"},"canonical_sha256":"0df1ecf7084228f7229846b8e2610d3c47610f9b152ee5756b5145ef0ef0fadd","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:18:54.465274Z","signature_b64":"xwwembS/3Ytqb8WPFyggroAgzBLRmT6OnT7EK6j6BjKBHE7dzp6RpccKLKvSu29THjeqGXA2LAfD5o8xL36jDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0df1ecf7084228f7229846b8e2610d3c47610f9b152ee5756b5145ef0ef0fadd","last_reissued_at":"2026-07-05T09:18:54.464794Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:18:54.464794Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2410.07471","source_version":2,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T09:18:54Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"ttl/U7Z1sWruMWxEYf+YmAuRod2WjsOd5qUBrqyfUKpB7WENN63Krs/vuHUUyD8notcwI1/hsr5fZ5xNESLPBw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-09T14:30:42.108027Z"},"content_sha256":"3dea1f6c66ba6510d33750a5532f206a0f9e15042a2a1ae27e654ca0fef7cdec","schema_version":"1.0","event_id":"sha256:3dea1f6c66ba6510d33750a5532f206a0f9e15042a2a1ae27e654ca0fef7cdec"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2024:BXY6Z5YIIIUPOIUYI24OEYINHR","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"SEAL: Safety-enhanced Aligned LLM Fine-tuning via Bilevel Data Selection","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Han Shen, Payel Das, Pin-Yu Chen, Tianyi Chen","submitted_at":"2024-10-09T22:24:22Z","abstract_excerpt":"Fine-tuning on task-specific data to boost downstream performance is a crucial step for leveraging Large Language Models (LLMs). However, previous studies have demonstrated that fine-tuning the models on several adversarial samples or even benign data can greatly comprise the model's pre-equipped alignment and safety capabilities. In this work, we propose SEAL, a novel framework to enhance safety in LLM fine-tuning. SEAL learns a data ranker based on the bilevel optimization to up rank the safe and high-quality fine-tuning data and down rank the unsafe or low-quality ones. Models trained with "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.07471","kind":"arxiv","version":2},"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/2410.07471/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"},"verdict_id":null},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T09:18:54Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"0QWD7dRNs9SDEU+KllQG5UmVLPQI2JN8tPpOJ7OJsQcSupYhv8Tt77UCNm+3m+t0kFdKf7Rq85dJ3nmviQbcBw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-09T14:30:42.108688Z"},"content_sha256":"9863927a45d83fd1bd1925926f206d22554610e07694e7880a100a349ea8f14b","schema_version":"1.0","event_id":"sha256:9863927a45d83fd1bd1925926f206d22554610e07694e7880a100a349ea8f14b"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/BXY6Z5YIIIUPOIUYI24OEYINHR/bundle.json","state_url":"https://pith.science/pith/BXY6Z5YIIIUPOIUYI24OEYINHR/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/BXY6Z5YIIIUPOIUYI24OEYINHR/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-08-09T14:30:42Z","links":{"resolver":"https://pith.science/pith/BXY6Z5YIIIUPOIUYI24OEYINHR","bundle":"https://pith.science/pith/BXY6Z5YIIIUPOIUYI24OEYINHR/bundle.json","state":"https://pith.science/pith/BXY6Z5YIIIUPOIUYI24OEYINHR/state.json","well_known_bundle":"https://pith.science/.well-known/pith/BXY6Z5YIIIUPOIUYI24OEYINHR/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2024:BXY6Z5YIIIUPOIUYI24OEYINHR","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"b51dceff7c7b916fa158865bcabf3830a8aa7a2438c1dfac665d497fb43f86d0","cross_cats_sorted":["cs.AI","cs.CL"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-10-09T22:24:22Z","title_canon_sha256":"2fcabf01881c1e39ab9b9a722376a470b0732eefd592682684fecc72fba29221"},"schema_version":"1.0","source":{"id":"2410.07471","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2410.07471","created_at":"2026-07-05T09:18:54Z"},{"alias_kind":"arxiv_version","alias_value":"2410.07471v2","created_at":"2026-07-05T09:18:54Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.07471","created_at":"2026-07-05T09:18:54Z"},{"alias_kind":"pith_short_12","alias_value":"BXY6Z5YIIIUP","created_at":"2026-07-05T09:18:54Z"},{"alias_kind":"pith_short_16","alias_value":"BXY6Z5YIIIUPOIUY","created_at":"2026-07-05T09:18:54Z"},{"alias_kind":"pith_short_8","alias_value":"BXY6Z5YI","created_at":"2026-07-05T09:18:54Z"}],"graph_snapshots":[{"event_id":"sha256:9863927a45d83fd1bd1925926f206d22554610e07694e7880a100a349ea8f14b","target":"graph","created_at":"2026-07-05T09:18:54Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2410.07471/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Fine-tuning on task-specific data to boost downstream performance is a crucial step for leveraging Large Language Models (LLMs). However, previous studies have demonstrated that fine-tuning the models on several adversarial samples or even benign data can greatly comprise the model's pre-equipped alignment and safety capabilities. In this work, we propose SEAL, a novel framework to enhance safety in LLM fine-tuning. SEAL learns a data ranker based on the bilevel optimization to up rank the safe and high-quality fine-tuning data and down rank the unsafe or low-quality ones. Models trained with ","authors_text":"Han Shen, Payel Das, Pin-Yu Chen, Tianyi Chen","cross_cats":["cs.AI","cs.CL"],"headline":"","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-10-09T22:24:22Z","title":"SEAL: Safety-enhanced Aligned LLM Fine-tuning via Bilevel Data Selection"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.07471","kind":"arxiv","version":2},"verdict":{"created_at":null,"id":null,"model_set":{},"one_line_summary":"","pipeline_version":null,"pith_extraction_headline":"","strongest_claim":"","weakest_assumption":""}},"verdict_id":null}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:3dea1f6c66ba6510d33750a5532f206a0f9e15042a2a1ae27e654ca0fef7cdec","target":"record","created_at":"2026-07-05T09:18:54Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"b51dceff7c7b916fa158865bcabf3830a8aa7a2438c1dfac665d497fb43f86d0","cross_cats_sorted":["cs.AI","cs.CL"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-10-09T22:24:22Z","title_canon_sha256":"2fcabf01881c1e39ab9b9a722376a470b0732eefd592682684fecc72fba29221"},"schema_version":"1.0","source":{"id":"2410.07471","kind":"arxiv","version":2}},"canonical_sha256":"0df1ecf7084228f7229846b8e2610d3c47610f9b152ee5756b5145ef0ef0fadd","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"0df1ecf7084228f7229846b8e2610d3c47610f9b152ee5756b5145ef0ef0fadd","first_computed_at":"2026-07-05T09:18:54.464794Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T09:18:54.464794Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"xwwembS/3Ytqb8WPFyggroAgzBLRmT6OnT7EK6j6BjKBHE7dzp6RpccKLKvSu29THjeqGXA2LAfD5o8xL36jDw==","signature_status":"signed_v1","signed_at":"2026-07-05T09:18:54.465274Z","signed_message":"canonical_sha256_bytes"},"source_id":"2410.07471","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:3dea1f6c66ba6510d33750a5532f206a0f9e15042a2a1ae27e654ca0fef7cdec","sha256:9863927a45d83fd1bd1925926f206d22554610e07694e7880a100a349ea8f14b"],"state_sha256":"f1b74c9371857d403d31dd6ef1ee3bc205971a3e4373e0d41b1a2b101d01b804"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"fSoeQowK1TopeBE0WqU9709/9qN+MeZklBCcsv/1QRZYAGixrb1vHzfYsa8WS5/zCECBYCJBxZQE0+3HhISrDA==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-09T14:30:42.114769Z","bundle_sha256":"aa9d99098dc1417d814078e52848e64b6de1730f28e58d2a1c25c8715b7a5888"}}