{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:ZHV6GX2556ABJWK2KIA5VXRMTS","short_pith_number":"pith:ZHV6GX25","schema_version":"1.0","canonical_sha256":"c9ebe35f5def8014d95a5201dade2c9cb65c075ea822c774dbc72c07876a970b","source":{"kind":"arxiv","id":"2405.18166","version":2},"attestation_state":"computed","paper":{"title":"Defending Large Language Models Against Jailbreak Attacks via Layer-specific Editing","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Jun Sun, Wei Zhao, Ye Zhang, Yige Li, Zhe Li","submitted_at":"2024-05-28T13:26:12Z","abstract_excerpt":"Large language models (LLMs) are increasingly being adopted in a wide range of real-world applications. Despite their impressive performance, recent studies have shown that LLMs are vulnerable to deliberately crafted adversarial prompts even when aligned via Reinforcement Learning from Human Feedback or supervised fine-tuning. While existing defense methods focus on either detecting harmful prompts or reducing the likelihood of harmful responses through various means, defending LLMs against jailbreak attacks based on the inner mechanisms of LLMs remains largely unexplored. In this work, we inv"},"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":"2405.18166","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2024-05-28T13:26:12Z","cross_cats_sorted":[],"title_canon_sha256":"ff1efc823a5f3c1af5b5964f6113277d674970d2aa5aa17e603dad74ce80d2e5","abstract_canon_sha256":"ef54ef1ee622f8f845db8a938babbcfdd5f88fb39397ef1ca59ec3c4943f1735"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:31:51.618919Z","signature_b64":"PgArI8emntHMuoVMoKtPlLeniwq7qQHh33slGg3ZH8qPOQIvabC3vvXwfavoRWi2rfd7QCVH88/821+D6dNbBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c9ebe35f5def8014d95a5201dade2c9cb65c075ea822c774dbc72c07876a970b","last_reissued_at":"2026-07-05T08:31:51.618421Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:31:51.618421Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Defending Large Language Models Against Jailbreak Attacks via Layer-specific Editing","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Jun Sun, Wei Zhao, Ye Zhang, Yige Li, Zhe Li","submitted_at":"2024-05-28T13:26:12Z","abstract_excerpt":"Large language models (LLMs) are increasingly being adopted in a wide range of real-world applications. Despite their impressive performance, recent studies have shown that LLMs are vulnerable to deliberately crafted adversarial prompts even when aligned via Reinforcement Learning from Human Feedback or supervised fine-tuning. While existing defense methods focus on either detecting harmful prompts or reducing the likelihood of harmful responses through various means, defending LLMs against jailbreak attacks based on the inner mechanisms of LLMs remains largely unexplored. In this work, we inv"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2405.18166","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/2405.18166/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":"2405.18166","created_at":"2026-07-05T08:31:51.618481+00:00"},{"alias_kind":"arxiv_version","alias_value":"2405.18166v2","created_at":"2026-07-05T08:31:51.618481+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2405.18166","created_at":"2026-07-05T08:31:51.618481+00:00"},{"alias_kind":"pith_short_12","alias_value":"ZHV6GX2556AB","created_at":"2026-07-05T08:31:51.618481+00:00"},{"alias_kind":"pith_short_16","alias_value":"ZHV6GX2556ABJWK2","created_at":"2026-07-05T08:31:51.618481+00:00"},{"alias_kind":"pith_short_8","alias_value":"ZHV6GX25","created_at":"2026-07-05T08:31:51.618481+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2506.00166","citing_title":"Disentangled Safety Adapters Enable Efficient Guardrails and Flexible Inference-Time Alignment","ref_index":53,"is_internal_anchor":false},{"citing_arxiv_id":"2512.12069","citing_title":"Rethinking Jailbreak Detection of Large Vision Language Models with Representational Contrastive Scoring","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2604.09544","citing_title":"Large Language Models Generate Harmful Responses Using a Distinct Mechanism, Shared Across Harm Types","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2604.07835","citing_title":"Silencing the Guardrails: Inference-Time Jailbreaking via Dynamic Contextual Representation Ablation","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2604.18519","citing_title":"LLM Safety From Within: Detecting Harmful Content with Internal Representations","ref_index":15,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ZHV6GX2556ABJWK2KIA5VXRMTS","json":"https://pith.science/pith/ZHV6GX2556ABJWK2KIA5VXRMTS.json","graph_json":"https://pith.science/api/pith-number/ZHV6GX2556ABJWK2KIA5VXRMTS/graph.json","events_json":"https://pith.science/api/pith-number/ZHV6GX2556ABJWK2KIA5VXRMTS/events.json","paper":"https://pith.science/paper/ZHV6GX25"},"agent_actions":{"view_html":"https://pith.science/pith/ZHV6GX2556ABJWK2KIA5VXRMTS","download_json":"https://pith.science/pith/ZHV6GX2556ABJWK2KIA5VXRMTS.json","view_paper":"https://pith.science/paper/ZHV6GX25","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2405.18166&json=true","fetch_graph":"https://pith.science/api/pith-number/ZHV6GX2556ABJWK2KIA5VXRMTS/graph.json","fetch_events":"https://pith.science/api/pith-number/ZHV6GX2556ABJWK2KIA5VXRMTS/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ZHV6GX2556ABJWK2KIA5VXRMTS/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ZHV6GX2556ABJWK2KIA5VXRMTS/action/storage_attestation","attest_author":"https://pith.science/pith/ZHV6GX2556ABJWK2KIA5VXRMTS/action/author_attestation","sign_citation":"https://pith.science/pith/ZHV6GX2556ABJWK2KIA5VXRMTS/action/citation_signature","submit_replication":"https://pith.science/pith/ZHV6GX2556ABJWK2KIA5VXRMTS/action/replication_record"}},"created_at":"2026-07-05T08:31:51.618481+00:00","updated_at":"2026-07-05T08:31:51.618481+00:00"}