{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:6BFNOLFL64UH3ZPYPGA3ST7GZU","short_pith_number":"pith:6BFNOLFL","schema_version":"1.0","canonical_sha256":"f04ad72cabf7287de5f87981b94fe6cd3d9fca9a7225c72acabb518e9a818ce8","source":{"kind":"arxiv","id":"2606.01738","version":1},"attestation_state":"computed","paper":{"title":"THRD: A Training-Free Multi-Turn Defense Framework for Jailbreak Attacks on Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Changliang Li, Chen Kang, Dong Yu, Pengyuan Liu, Zhiqing Ma, Zhonghao Xu","submitted_at":"2026-06-01T06:02:15Z","abstract_excerpt":"Multi-turn jailbreak attacks pose a growing threat to LLMs by exploiting conversational dynamics such as gradual escalation and cross-turn coordination. Existing defenses either rely on costly retraining -- often degrading model utility -- or apply single-turn analysis independently at each turn, failing to capture how risk accumulates along interaction trajectories. We observe that safety behavior in multi-turn interaction is trajectory-dependent: dialogue history continuously reshapes the model's conditioning context, making it insufficient to evaluate each turn in isolation. Motivated by th"},"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":"2606.01738","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-06-01T06:02:15Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"af096b21b8f10c9d5492c8e9c1a9368d750489f77567cb4da314f43bba491cf7","abstract_canon_sha256":"9a38dbe96e76e257e06311f0b0c53f510f053617393adf30666756610c181634"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-02T02:04:55.464329Z","signature_b64":"oz+7kvxBzptJxVPqhth7UBvZ6LJe+vcBoegEzxMojmOn2sofxUgq331GBBWFRxdM9Qj2dvF7+LWqzYsku67fBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f04ad72cabf7287de5f87981b94fe6cd3d9fca9a7225c72acabb518e9a818ce8","last_reissued_at":"2026-06-02T02:04:55.463962Z","signature_status":"signed_v1","first_computed_at":"2026-06-02T02:04:55.463962Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"THRD: A Training-Free Multi-Turn Defense Framework for Jailbreak Attacks on Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Changliang Li, Chen Kang, Dong Yu, Pengyuan Liu, Zhiqing Ma, Zhonghao Xu","submitted_at":"2026-06-01T06:02:15Z","abstract_excerpt":"Multi-turn jailbreak attacks pose a growing threat to LLMs by exploiting conversational dynamics such as gradual escalation and cross-turn coordination. Existing defenses either rely on costly retraining -- often degrading model utility -- or apply single-turn analysis independently at each turn, failing to capture how risk accumulates along interaction trajectories. We observe that safety behavior in multi-turn interaction is trajectory-dependent: dialogue history continuously reshapes the model's conditioning context, making it insufficient to evaluate each turn in isolation. Motivated by th"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.01738","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/2606.01738/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":"2606.01738","created_at":"2026-06-02T02:04:55.464020+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.01738v1","created_at":"2026-06-02T02:04:55.464020+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.01738","created_at":"2026-06-02T02:04:55.464020+00:00"},{"alias_kind":"pith_short_12","alias_value":"6BFNOLFL64UH","created_at":"2026-06-02T02:04:55.464020+00:00"},{"alias_kind":"pith_short_16","alias_value":"6BFNOLFL64UH3ZPY","created_at":"2026-06-02T02:04:55.464020+00:00"},{"alias_kind":"pith_short_8","alias_value":"6BFNOLFL","created_at":"2026-06-02T02:04:55.464020+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/6BFNOLFL64UH3ZPYPGA3ST7GZU","json":"https://pith.science/pith/6BFNOLFL64UH3ZPYPGA3ST7GZU.json","graph_json":"https://pith.science/api/pith-number/6BFNOLFL64UH3ZPYPGA3ST7GZU/graph.json","events_json":"https://pith.science/api/pith-number/6BFNOLFL64UH3ZPYPGA3ST7GZU/events.json","paper":"https://pith.science/paper/6BFNOLFL"},"agent_actions":{"view_html":"https://pith.science/pith/6BFNOLFL64UH3ZPYPGA3ST7GZU","download_json":"https://pith.science/pith/6BFNOLFL64UH3ZPYPGA3ST7GZU.json","view_paper":"https://pith.science/paper/6BFNOLFL","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.01738&json=true","fetch_graph":"https://pith.science/api/pith-number/6BFNOLFL64UH3ZPYPGA3ST7GZU/graph.json","fetch_events":"https://pith.science/api/pith-number/6BFNOLFL64UH3ZPYPGA3ST7GZU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/6BFNOLFL64UH3ZPYPGA3ST7GZU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/6BFNOLFL64UH3ZPYPGA3ST7GZU/action/storage_attestation","attest_author":"https://pith.science/pith/6BFNOLFL64UH3ZPYPGA3ST7GZU/action/author_attestation","sign_citation":"https://pith.science/pith/6BFNOLFL64UH3ZPYPGA3ST7GZU/action/citation_signature","submit_replication":"https://pith.science/pith/6BFNOLFL64UH3ZPYPGA3ST7GZU/action/replication_record"}},"created_at":"2026-06-02T02:04:55.464020+00:00","updated_at":"2026-06-02T02:04:55.464020+00:00"}