{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2025:5MXN5A4J2ZP2XWPPT7ON5LIR6H","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":"fa6161b8aa84aed8c4d58573058e97e0d538bce6bbce2ca9f84a259d90c85bac","cross_cats_sorted":["cs.CL"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-05-11T04:59:20Z","title_canon_sha256":"b45e6e704ab925922e0e40e131f6f65f38f28219cdd94b6890e92bd3acf33438"},"schema_version":"1.0","source":{"id":"2505.06843","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2505.06843","created_at":"2026-07-05T11:09:13Z"},{"alias_kind":"arxiv_version","alias_value":"2505.06843v2","created_at":"2026-07-05T11:09:13Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.06843","created_at":"2026-07-05T11:09:13Z"},{"alias_kind":"pith_short_12","alias_value":"5MXN5A4J2ZP2","created_at":"2026-07-05T11:09:13Z"},{"alias_kind":"pith_short_16","alias_value":"5MXN5A4J2ZP2XWPP","created_at":"2026-07-05T11:09:13Z"},{"alias_kind":"pith_short_8","alias_value":"5MXN5A4J","created_at":"2026-07-05T11:09:13Z"}],"graph_snapshots":[{"event_id":"sha256:1e07bc5b0c91feeda12aca3d5cefa78415d99f529289213370c0503cc87f57b7","target":"graph","created_at":"2026-07-05T11:09:13Z","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/2505.06843/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Recent studies have uncovered a troubling vulnerability in the fine-tuning stage of large language models (LLMs): even fine-tuning on entirely benign datasets can lead to a significant increase in the harmfulness of LLM outputs. Building on this finding, our red teaming study takes this threat one step further by developing a more effective attack. Specifically, we analyze and identify samples within benign datasets that contribute most to safety degradation, then fine-tune LLMs exclusively on these samples. We approach this problem from an outlier detection perspective and propose Self-Inf-N,","authors_text":"Anil Vullikanti, Mengxuan Hu, Ronghang Zhu, Sheng Li, Zihan Guan","cross_cats":["cs.CL"],"headline":"","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-05-11T04:59:20Z","title":"Benign Samples Matter! Fine-tuning On Outlier Benign Samples Severely Breaks Safety"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.06843","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":"sha256:8e0e3cfb1b1b4bf7b3e7c6df664a16fd64c1765ebff7999e04107c80233273d5","target":"record","created_at":"2026-07-05T11:09:13Z","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":"fa6161b8aa84aed8c4d58573058e97e0d538bce6bbce2ca9f84a259d90c85bac","cross_cats_sorted":["cs.CL"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-05-11T04:59:20Z","title_canon_sha256":"b45e6e704ab925922e0e40e131f6f65f38f28219cdd94b6890e92bd3acf33438"},"schema_version":"1.0","source":{"id":"2505.06843","kind":"arxiv","version":2}},"canonical_sha256":"eb2ede8389d65fabd9ef9fdcdead11f1c2e38f0ea0393e794043ae9105ddb40d","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"eb2ede8389d65fabd9ef9fdcdead11f1c2e38f0ea0393e794043ae9105ddb40d","first_computed_at":"2026-07-05T11:09:13.079971Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T11:09:13.079971Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"tD90/21edIQ8OFT6As8TDC9REC4DfAVvyVbiiDBoK8fQ8bUmmjybHmm9u8qhzkjato07NanNMcveGPePq/uLBA==","signature_status":"signed_v1","signed_at":"2026-07-05T11:09:13.080469Z","signed_message":"canonical_sha256_bytes"},"source_id":"2505.06843","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:8e0e3cfb1b1b4bf7b3e7c6df664a16fd64c1765ebff7999e04107c80233273d5","sha256:1e07bc5b0c91feeda12aca3d5cefa78415d99f529289213370c0503cc87f57b7"],"state_sha256":"ae21ff9d81062cf2021eb11d84e6eb56d59713d84d6950aadc3af75e8baa0247"}