{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2024:E7H2S6GTCMUS3O4GSMU2FTUPRL","short_pith_number":"pith:E7H2S6GT","canonical_record":{"source":{"id":"2408.17003","kind":"arxiv","version":5},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2024-08-30T04:35:59Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"aae438b7dc71a951e1600fc129acd7be4f8b51cbb522ec305410deaf2aa0c66b","abstract_canon_sha256":"72c267d07a8b897b48768a4635629556d9f4a24fcf529eb0fec3ed35c48ca942"},"schema_version":"1.0"},"canonical_sha256":"27cfa978d313292dbb869329a2ce8f8ad192f9b6bf26e3dd5ce23054805b92ba","source":{"kind":"arxiv","id":"2408.17003","version":5},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2408.17003","created_at":"2026-07-05T10:45:11Z"},{"alias_kind":"arxiv_version","alias_value":"2408.17003v5","created_at":"2026-07-05T10:45:11Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2408.17003","created_at":"2026-07-05T10:45:11Z"},{"alias_kind":"pith_short_12","alias_value":"E7H2S6GTCMUS","created_at":"2026-07-05T10:45:11Z"},{"alias_kind":"pith_short_16","alias_value":"E7H2S6GTCMUS3O4G","created_at":"2026-07-05T10:45:11Z"},{"alias_kind":"pith_short_8","alias_value":"E7H2S6GT","created_at":"2026-07-05T10:45:11Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2024:E7H2S6GTCMUS3O4GSMU2FTUPRL","target":"record","payload":{"canonical_record":{"source":{"id":"2408.17003","kind":"arxiv","version":5},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2024-08-30T04:35:59Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"aae438b7dc71a951e1600fc129acd7be4f8b51cbb522ec305410deaf2aa0c66b","abstract_canon_sha256":"72c267d07a8b897b48768a4635629556d9f4a24fcf529eb0fec3ed35c48ca942"},"schema_version":"1.0"},"canonical_sha256":"27cfa978d313292dbb869329a2ce8f8ad192f9b6bf26e3dd5ce23054805b92ba","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:45:11.125120Z","signature_b64":"+XYOdlOvKDuIzTIiDY30pRaqPbkPJXYPwbkyvrKmgxCrzwMs/mIbwuK2YBBpc2SK/yRw8YJL5r1WUGg2Y8JGBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"27cfa978d313292dbb869329a2ce8f8ad192f9b6bf26e3dd5ce23054805b92ba","last_reissued_at":"2026-07-05T10:45:11.124584Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:45:11.124584Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2408.17003","source_version":5,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T10:45:11Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"95Rf6x9EIM/VF6KlHBjoEQp52i70txpKAI8foUIJRgbS1Bj0clWvtX4DvMEDkxPZUhjfOTbXIN9DjhSC8TBLBA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-07T19:17:45.101175Z"},"content_sha256":"0deabece5f26df12c869cb707d5a27e78f19409a3a4c15be7e24e184b08ada3c","schema_version":"1.0","event_id":"sha256:0deabece5f26df12c869cb707d5a27e78f19409a3a4c15be7e24e184b08ada3c"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2024:E7H2S6GTCMUS3O4GSMU2FTUPRL","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Safety Layers in Aligned Large Language Models: The Key to LLM Security","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CR","authors_text":"Lan Zhang, Liuyi Yao, Shen Li, Yaliang Li","submitted_at":"2024-08-30T04:35:59Z","abstract_excerpt":"Aligned LLMs are secure, capable of recognizing and refusing to answer malicious questions. However, the role of internal parameters in maintaining such security is not well understood yet, further these models can be vulnerable to security degradation when subjected to fine-tuning attacks. To address these challenges, our work uncovers the mechanism behind security in aligned LLMs at the parameter level, identifying a small set of contiguous layers in the middle of the model that are crucial for distinguishing malicious queries from normal ones, referred to as ``safety layers\". We first confi"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2408.17003","kind":"arxiv","version":5},"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/2408.17003/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":"2026-07-05T10:45:11Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"gj0/kNl0rHyb3lVoLK9s/kT556JgpqlMtEcJ14zuxM/FMevz81EOhX2dd5uAq+vqH/vV7MVzuSr2PyizEyyYAg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-07T19:17:45.101828Z"},"content_sha256":"211d7c3f686f71e71a76f5915d832027fbe8f744ff5fe870fa392d08c1112a48","schema_version":"1.0","event_id":"sha256:211d7c3f686f71e71a76f5915d832027fbe8f744ff5fe870fa392d08c1112a48"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/E7H2S6GTCMUS3O4GSMU2FTUPRL/bundle.json","state_url":"https://pith.science/pith/E7H2S6GTCMUS3O4GSMU2FTUPRL/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/E7H2S6GTCMUS3O4GSMU2FTUPRL/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":"2026-08-07T19:17:45Z","links":{"resolver":"https://pith.science/pith/E7H2S6GTCMUS3O4GSMU2FTUPRL","bundle":"https://pith.science/pith/E7H2S6GTCMUS3O4GSMU2FTUPRL/bundle.json","state":"https://pith.science/pith/E7H2S6GTCMUS3O4GSMU2FTUPRL/state.json","well_known_bundle":"https://pith.science/.well-known/pith/E7H2S6GTCMUS3O4GSMU2FTUPRL/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2024:E7H2S6GTCMUS3O4GSMU2FTUPRL","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":"72c267d07a8b897b48768a4635629556d9f4a24fcf529eb0fec3ed35c48ca942","cross_cats_sorted":["cs.AI"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2024-08-30T04:35:59Z","title_canon_sha256":"aae438b7dc71a951e1600fc129acd7be4f8b51cbb522ec305410deaf2aa0c66b"},"schema_version":"1.0","source":{"id":"2408.17003","kind":"arxiv","version":5}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2408.17003","created_at":"2026-07-05T10:45:11Z"},{"alias_kind":"arxiv_version","alias_value":"2408.17003v5","created_at":"2026-07-05T10:45:11Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2408.17003","created_at":"2026-07-05T10:45:11Z"},{"alias_kind":"pith_short_12","alias_value":"E7H2S6GTCMUS","created_at":"2026-07-05T10:45:11Z"},{"alias_kind":"pith_short_16","alias_value":"E7H2S6GTCMUS3O4G","created_at":"2026-07-05T10:45:11Z"},{"alias_kind":"pith_short_8","alias_value":"E7H2S6GT","created_at":"2026-07-05T10:45:11Z"}],"graph_snapshots":[{"event_id":"sha256:211d7c3f686f71e71a76f5915d832027fbe8f744ff5fe870fa392d08c1112a48","target":"graph","created_at":"2026-07-05T10:45:11Z","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/2408.17003/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Aligned LLMs are secure, capable of recognizing and refusing to answer malicious questions. However, the role of internal parameters in maintaining such security is not well understood yet, further these models can be vulnerable to security degradation when subjected to fine-tuning attacks. To address these challenges, our work uncovers the mechanism behind security in aligned LLMs at the parameter level, identifying a small set of contiguous layers in the middle of the model that are crucial for distinguishing malicious queries from normal ones, referred to as ``safety layers\". We first confi","authors_text":"Lan Zhang, Liuyi Yao, Shen Li, Yaliang Li","cross_cats":["cs.AI"],"headline":"","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2024-08-30T04:35:59Z","title":"Safety Layers in Aligned Large Language Models: The Key to LLM Security"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2408.17003","kind":"arxiv","version":5},"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":"sha256:0deabece5f26df12c869cb707d5a27e78f19409a3a4c15be7e24e184b08ada3c","target":"record","created_at":"2026-07-05T10:45:11Z","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":"72c267d07a8b897b48768a4635629556d9f4a24fcf529eb0fec3ed35c48ca942","cross_cats_sorted":["cs.AI"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2024-08-30T04:35:59Z","title_canon_sha256":"aae438b7dc71a951e1600fc129acd7be4f8b51cbb522ec305410deaf2aa0c66b"},"schema_version":"1.0","source":{"id":"2408.17003","kind":"arxiv","version":5}},"canonical_sha256":"27cfa978d313292dbb869329a2ce8f8ad192f9b6bf26e3dd5ce23054805b92ba","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"27cfa978d313292dbb869329a2ce8f8ad192f9b6bf26e3dd5ce23054805b92ba","first_computed_at":"2026-07-05T10:45:11.124584Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T10:45:11.124584Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"+XYOdlOvKDuIzTIiDY30pRaqPbkPJXYPwbkyvrKmgxCrzwMs/mIbwuK2YBBpc2SK/yRw8YJL5r1WUGg2Y8JGBQ==","signature_status":"signed_v1","signed_at":"2026-07-05T10:45:11.125120Z","signed_message":"canonical_sha256_bytes"},"source_id":"2408.17003","source_kind":"arxiv","source_version":5}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:0deabece5f26df12c869cb707d5a27e78f19409a3a4c15be7e24e184b08ada3c","sha256:211d7c3f686f71e71a76f5915d832027fbe8f744ff5fe870fa392d08c1112a48"],"state_sha256":"dbbc527acf59a2ebc8b7e30cd6916ff0bbbee9c48f657d2e2c6359ca67cd53fb"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"ZIWDnADfieX3f+wQh32knLiuHEuLhdeKyDpcJum6h9e3sDzgQDJpmwvmd8YtOAnhAJEiPWBczIb93hoI5hPsAg==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-07T19:17:45.108569Z","bundle_sha256":"22278c1c6d49c31c4de02ea5fb0538d9e8adb1d9dd86c2ea197e8b02abe69c61"}}