{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2024:ODALYDF2OTUDXWHQZMHNZY4FPO","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":"54acf59667ff6ccfac9816ddab5f5c5d91785bff227fa2d91c6e6d70cf494667","cross_cats_sorted":["cs.AI","cs.CL"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-02-07T18:34:38Z","title_canon_sha256":"3344489afdccf9f16e936a88ebd47396f69ae75967ee7a9a707dab0cb3e721b4"},"schema_version":"1.0","source":{"id":"2402.05162","kind":"arxiv","version":4}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2402.05162","created_at":"2026-07-05T09:25:39Z"},{"alias_kind":"arxiv_version","alias_value":"2402.05162v4","created_at":"2026-07-05T09:25:39Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2402.05162","created_at":"2026-07-05T09:25:39Z"},{"alias_kind":"pith_short_12","alias_value":"ODALYDF2OTUD","created_at":"2026-07-05T09:25:39Z"},{"alias_kind":"pith_short_16","alias_value":"ODALYDF2OTUDXWHQ","created_at":"2026-07-05T09:25:39Z"},{"alias_kind":"pith_short_8","alias_value":"ODALYDF2","created_at":"2026-07-05T09:25:39Z"}],"graph_snapshots":[{"event_id":"sha256:be724f5e90afe8f644e3b43098db7a75ae4298c8615c7392c38ae29c804cd3a5","target":"graph","created_at":"2026-07-05T09:25:39Z","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/2402.05162/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Large language models (LLMs) show inherent brittleness in their safety mechanisms, as evidenced by their susceptibility to jailbreaking and even non-malicious fine-tuning. This study explores this brittleness of safety alignment by leveraging pruning and low-rank modifications. We develop methods to identify critical regions that are vital for safety guardrails, and that are disentangled from utility-relevant regions at both the neuron and rank levels. Surprisingly, the isolated regions we find are sparse, comprising about $3\\%$ at the parameter level and $2.5\\%$ at the rank level. Removing th","authors_text":"Boyi Wei, Kaixuan Huang, Mengdi Wang, Mengzhou Xia, Peter Henderson, Prateek Mittal, Tinghao Xie, Xiangyu Qi, Yangsibo Huang","cross_cats":["cs.AI","cs.CL"],"headline":"","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-02-07T18:34:38Z","title":"Assessing the Brittleness of Safety Alignment via Pruning and Low-Rank Modifications"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2402.05162","kind":"arxiv","version":4},"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":"sha256:91e55cd19433177d8c5269103baa2279929add3f9e99b8c4c325f2005f6b475d","target":"record","created_at":"2026-07-05T09:25:39Z","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":"54acf59667ff6ccfac9816ddab5f5c5d91785bff227fa2d91c6e6d70cf494667","cross_cats_sorted":["cs.AI","cs.CL"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-02-07T18:34:38Z","title_canon_sha256":"3344489afdccf9f16e936a88ebd47396f69ae75967ee7a9a707dab0cb3e721b4"},"schema_version":"1.0","source":{"id":"2402.05162","kind":"arxiv","version":4}},"canonical_sha256":"70c0bc0cba74e83bd8f0cb0edce3857ba0f99944f5681faf384d7bf7e7411d9c","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"70c0bc0cba74e83bd8f0cb0edce3857ba0f99944f5681faf384d7bf7e7411d9c","first_computed_at":"2026-07-05T09:25:39.490118Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T09:25:39.490118Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"26B7orpxxgJZtiq0zHebfYh/xaN0yBUIuk3FuIXWsDo6Qln0qhHXUSUVZoXk1shu+/aBv/eJV7OdX34szhnABQ==","signature_status":"signed_v1","signed_at":"2026-07-05T09:25:39.490603Z","signed_message":"canonical_sha256_bytes"},"source_id":"2402.05162","source_kind":"arxiv","source_version":4}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:91e55cd19433177d8c5269103baa2279929add3f9e99b8c4c325f2005f6b475d","sha256:be724f5e90afe8f644e3b43098db7a75ae4298c8615c7392c38ae29c804cd3a5"],"state_sha256":"6801f5ae7326896fb32d035492f218d64d0bb9dba7633972d49ab12d184c0017"}