{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:HL3WGLCYM3257BRM3CZGI3PHKM","short_pith_number":"pith:HL3WGLCY","schema_version":"1.0","canonical_sha256":"3af7632c5866f5df862cd8b2646de75300ef93dd4118cef64a21f3044af40387","source":{"kind":"arxiv","id":"2606.07335","version":1},"attestation_state":"computed","paper":{"title":"Defending Jailbreak Attacks on Large Language Models via Manifold Trajectory Kinetics","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CR","authors_text":"Hangtao Zhang, Leo Yu Zhang, Minghui Li, Shengshan Hu, Sishun Liu, Wei Wan, Yanjun Zhang, Yi Liu, Yucheng Zhao, Zeyu Ye, Ziqi Zhou","submitted_at":"2026-06-05T14:49:26Z","abstract_excerpt":"Jailbreak prompts can bypass alignment guardrails in large language models (LLMs) and elicit unsafe outputs, making reliable deployment-time detection critical. Prior detection approaches largely rely on a fixed metric space, e.g., raw inputs, gradients, or hidden features, in which benign and jailbreak prompts are linearly separable. We show this assumption breaks under (i) pseudo-malicious prompts that are benign by intent but contain safety-related keywords, and (ii) adaptive attacks that explicitly optimize against the deployed detector. To overcome this limitation, we shift our focus from"},"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":"2606.07335","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2026-06-05T14:49:26Z","cross_cats_sorted":[],"title_canon_sha256":"2685cd61bc6052a1d6b9fe6010de49bbed4301253e41c40ff2ffd66bed641cea","abstract_canon_sha256":"16c48fbdbab7339f3b120c67ea97ff12a042d9b8c55c86a6f52e1ca2a3ee32f5"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-08T01:05:20.397330Z","signature_b64":"s3R0P4mJ5iOEtOcWbJrnaEdw0C/X4c71C/KkHekvEBTwR6e5GHV3vOPykCDIRuGmFddtVEMQwa81vEdUj9EdCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"3af7632c5866f5df862cd8b2646de75300ef93dd4118cef64a21f3044af40387","last_reissued_at":"2026-06-08T01:05:20.396572Z","signature_status":"signed_v1","first_computed_at":"2026-06-08T01:05:20.396572Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Defending Jailbreak Attacks on Large Language Models via Manifold Trajectory Kinetics","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CR","authors_text":"Hangtao Zhang, Leo Yu Zhang, Minghui Li, Shengshan Hu, Sishun Liu, Wei Wan, Yanjun Zhang, Yi Liu, Yucheng Zhao, Zeyu Ye, Ziqi Zhou","submitted_at":"2026-06-05T14:49:26Z","abstract_excerpt":"Jailbreak prompts can bypass alignment guardrails in large language models (LLMs) and elicit unsafe outputs, making reliable deployment-time detection critical. Prior detection approaches largely rely on a fixed metric space, e.g., raw inputs, gradients, or hidden features, in which benign and jailbreak prompts are linearly separable. We show this assumption breaks under (i) pseudo-malicious prompts that are benign by intent but contain safety-related keywords, and (ii) adaptive attacks that explicitly optimize against the deployed detector. To overcome this limitation, we shift our focus from"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.07335","kind":"arxiv","version":1},"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/2606.07335/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":"2606.07335","created_at":"2026-06-08T01:05:20.396702+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.07335v1","created_at":"2026-06-08T01:05:20.396702+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.07335","created_at":"2026-06-08T01:05:20.396702+00:00"},{"alias_kind":"pith_short_12","alias_value":"HL3WGLCYM325","created_at":"2026-06-08T01:05:20.396702+00:00"},{"alias_kind":"pith_short_16","alias_value":"HL3WGLCYM3257BRM","created_at":"2026-06-08T01:05:20.396702+00:00"},{"alias_kind":"pith_short_8","alias_value":"HL3WGLCY","created_at":"2026-06-08T01:05:20.396702+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/HL3WGLCYM3257BRM3CZGI3PHKM","json":"https://pith.science/pith/HL3WGLCYM3257BRM3CZGI3PHKM.json","graph_json":"https://pith.science/api/pith-number/HL3WGLCYM3257BRM3CZGI3PHKM/graph.json","events_json":"https://pith.science/api/pith-number/HL3WGLCYM3257BRM3CZGI3PHKM/events.json","paper":"https://pith.science/paper/HL3WGLCY"},"agent_actions":{"view_html":"https://pith.science/pith/HL3WGLCYM3257BRM3CZGI3PHKM","download_json":"https://pith.science/pith/HL3WGLCYM3257BRM3CZGI3PHKM.json","view_paper":"https://pith.science/paper/HL3WGLCY","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.07335&json=true","fetch_graph":"https://pith.science/api/pith-number/HL3WGLCYM3257BRM3CZGI3PHKM/graph.json","fetch_events":"https://pith.science/api/pith-number/HL3WGLCYM3257BRM3CZGI3PHKM/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/HL3WGLCYM3257BRM3CZGI3PHKM/action/timestamp_anchor","attest_storage":"https://pith.science/pith/HL3WGLCYM3257BRM3CZGI3PHKM/action/storage_attestation","attest_author":"https://pith.science/pith/HL3WGLCYM3257BRM3CZGI3PHKM/action/author_attestation","sign_citation":"https://pith.science/pith/HL3WGLCYM3257BRM3CZGI3PHKM/action/citation_signature","submit_replication":"https://pith.science/pith/HL3WGLCYM3257BRM3CZGI3PHKM/action/replication_record"}},"created_at":"2026-06-08T01:05:20.396702+00:00","updated_at":"2026-06-08T01:05:20.396702+00:00"}