{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:YH7RP5Y2PSHCOUFLGJXJDUEA34","short_pith_number":"pith:YH7RP5Y2","schema_version":"1.0","canonical_sha256":"c1ff17f71a7c8e2750ab326e91d080df10c431860ddd90826b64f79213beab59","source":{"kind":"arxiv","id":"2403.00867","version":3},"attestation_state":"computed","paper":{"title":"Gradient Cuff: Detecting Jailbreak Attacks on Large Language Models by Exploring Refusal Loss Landscapes","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.LG"],"primary_cat":"cs.CR","authors_text":"Pin-Yu Chen, Tsung-Yi Ho, Xiaomeng Hu","submitted_at":"2024-03-01T03:29:54Z","abstract_excerpt":"Large Language Models (LLMs) are becoming a prominent generative AI tool, where the user enters a query and the LLM generates an answer. To reduce harm and misuse, efforts have been made to align these LLMs to human values using advanced training techniques such as Reinforcement Learning from Human Feedback (RLHF). However, recent studies have highlighted the vulnerability of LLMs to adversarial jailbreak attempts aiming at subverting the embedded safety guardrails. To address this challenge, this paper defines and investigates the Refusal Loss of LLMs and then proposes a method called Gradien"},"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":"2403.00867","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2024-03-01T03:29:54Z","cross_cats_sorted":["cs.AI","cs.CL","cs.LG"],"title_canon_sha256":"b6a7d4d270430dfaf03c230930c90290514baa6ffbe3a481c1aa0ab70f69710e","abstract_canon_sha256":"5a561f4a8aed206885ba439b8e8473ccc39884546a4ccdf5fe5273e230210438"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:32:31.534751Z","signature_b64":"4PjCdvlq/eWrXAnoQ4EMrpckgWhoQW0O36Paz4jqiEHbFvRN1lXOB5zxuEwrcXJfeSvYf3tDWWlsMFq86kyPBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c1ff17f71a7c8e2750ab326e91d080df10c431860ddd90826b64f79213beab59","last_reissued_at":"2026-07-05T09:32:31.534236Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:32:31.534236Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Gradient Cuff: Detecting Jailbreak Attacks on Large Language Models by Exploring Refusal Loss Landscapes","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.LG"],"primary_cat":"cs.CR","authors_text":"Pin-Yu Chen, Tsung-Yi Ho, Xiaomeng Hu","submitted_at":"2024-03-01T03:29:54Z","abstract_excerpt":"Large Language Models (LLMs) are becoming a prominent generative AI tool, where the user enters a query and the LLM generates an answer. To reduce harm and misuse, efforts have been made to align these LLMs to human values using advanced training techniques such as Reinforcement Learning from Human Feedback (RLHF). However, recent studies have highlighted the vulnerability of LLMs to adversarial jailbreak attempts aiming at subverting the embedded safety guardrails. To address this challenge, this paper defines and investigates the Refusal Loss of LLMs and then proposes a method called Gradien"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2403.00867","kind":"arxiv","version":3},"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/2403.00867/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":"2403.00867","created_at":"2026-07-05T09:32:31.534289+00:00"},{"alias_kind":"arxiv_version","alias_value":"2403.00867v3","created_at":"2026-07-05T09:32:31.534289+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2403.00867","created_at":"2026-07-05T09:32:31.534289+00:00"},{"alias_kind":"pith_short_12","alias_value":"YH7RP5Y2PSHC","created_at":"2026-07-05T09:32:31.534289+00:00"},{"alias_kind":"pith_short_16","alias_value":"YH7RP5Y2PSHCOUFL","created_at":"2026-07-05T09:32:31.534289+00:00"},{"alias_kind":"pith_short_8","alias_value":"YH7RP5Y2","created_at":"2026-07-05T09:32:31.534289+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2504.00446","citing_title":"Exposing the Ghost in the Transformer: Abnormal Detection for Large Language Models via Hidden State Forensics","ref_index":43,"is_internal_anchor":false},{"citing_arxiv_id":"2508.04204","citing_title":"ReasoningGuard: Safeguarding Large Reasoning Models with Inference-time Safety Aha Moments","ref_index":40,"is_internal_anchor":false},{"citing_arxiv_id":"2407.04295","citing_title":"Jailbreak Attacks and Defenses Against Large Language Models: A Survey","ref_index":35,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/YH7RP5Y2PSHCOUFLGJXJDUEA34","json":"https://pith.science/pith/YH7RP5Y2PSHCOUFLGJXJDUEA34.json","graph_json":"https://pith.science/api/pith-number/YH7RP5Y2PSHCOUFLGJXJDUEA34/graph.json","events_json":"https://pith.science/api/pith-number/YH7RP5Y2PSHCOUFLGJXJDUEA34/events.json","paper":"https://pith.science/paper/YH7RP5Y2"},"agent_actions":{"view_html":"https://pith.science/pith/YH7RP5Y2PSHCOUFLGJXJDUEA34","download_json":"https://pith.science/pith/YH7RP5Y2PSHCOUFLGJXJDUEA34.json","view_paper":"https://pith.science/paper/YH7RP5Y2","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2403.00867&json=true","fetch_graph":"https://pith.science/api/pith-number/YH7RP5Y2PSHCOUFLGJXJDUEA34/graph.json","fetch_events":"https://pith.science/api/pith-number/YH7RP5Y2PSHCOUFLGJXJDUEA34/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YH7RP5Y2PSHCOUFLGJXJDUEA34/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YH7RP5Y2PSHCOUFLGJXJDUEA34/action/storage_attestation","attest_author":"https://pith.science/pith/YH7RP5Y2PSHCOUFLGJXJDUEA34/action/author_attestation","sign_citation":"https://pith.science/pith/YH7RP5Y2PSHCOUFLGJXJDUEA34/action/citation_signature","submit_replication":"https://pith.science/pith/YH7RP5Y2PSHCOUFLGJXJDUEA34/action/replication_record"}},"created_at":"2026-07-05T09:32:31.534289+00:00","updated_at":"2026-07-05T09:32:31.534289+00:00"}