{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:OMX2R4TNZI2U4I2SRFOVNQUDL4","short_pith_number":"pith:OMX2R4TN","schema_version":"1.0","canonical_sha256":"732fa8f26dca354e2352895d56c2835f11e28dc6c213fe8614ae0e62cccc9f1c","source":{"kind":"arxiv","id":"2603.15684","version":2},"attestation_state":"computed","paper":{"title":"State-Dependent Safety Failures in Multi-Turn Language Model Interaction","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CR","authors_text":"Han Qiu, Jie Zhang, Nenghai Yu, Pengcheng Li, Tianwei Zhang, Weiming Zhang, Wenbo Zhou, Zhang kejun","submitted_at":"2026-03-15T12:13:01Z","abstract_excerpt":"Safety alignment in large language models is typically evaluated under isolated queries, yet real-world use is inherently multi-turn. Although multi-turn jailbreaks are empirically effective, the structure of conversational safety failure remains insufficiently understood. In this work, we study safety failures from a state-space perspective and show that many multi-turn safety failures in current safety-aligned language models arise from contextual state evolution, a regime that is not fully captured by isolated prompt-level analyses alone. We introduce STAR, a state-oriented diagnostic frame"},"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":"2603.15684","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2026-03-15T12:13:01Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"280c609a2f7cc945aba8a873b29a485d754ce7073c43ee521462cf596734b782","abstract_canon_sha256":"a007b43d5a3f7e1edc38e140b268bff61a4d59a651c79b08ae42f9e87cc4f8d3"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"732fa8f26dca354e2352895d56c2835f11e28dc6c213fe8614ae0e62cccc9f1c","last_reissued_at":"2026-07-30T01:19:41.570042Z","signature_status":"unsigned_v0","first_computed_at":"2026-07-30T01:19:41.570042Z"},"graph_snapshot":{"paper":{"title":"State-Dependent Safety Failures in Multi-Turn Language Model Interaction","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CR","authors_text":"Han Qiu, Jie Zhang, Nenghai Yu, Pengcheng Li, Tianwei Zhang, Weiming Zhang, Wenbo Zhou, Zhang kejun","submitted_at":"2026-03-15T12:13:01Z","abstract_excerpt":"Safety alignment in large language models is typically evaluated under isolated queries, yet real-world use is inherently multi-turn. Although multi-turn jailbreaks are empirically effective, the structure of conversational safety failure remains insufficiently understood. In this work, we study safety failures from a state-space perspective and show that many multi-turn safety failures in current safety-aligned language models arise from contextual state evolution, a regime that is not fully captured by isolated prompt-level analyses alone. We introduce STAR, a state-oriented diagnostic frame"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2603.15684","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/2603.15684/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":"2603.15684","created_at":"2026-07-30T01:19:41.575114+00:00"},{"alias_kind":"arxiv_version","alias_value":"2603.15684v2","created_at":"2026-07-30T01:19:41.575114+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2603.15684","created_at":"2026-07-30T01:19:41.575114+00:00"},{"alias_kind":"pith_short_12","alias_value":"OMX2R4TNZI2U","created_at":"2026-07-30T01:19:41.575114+00:00"},{"alias_kind":"pith_short_16","alias_value":"OMX2R4TNZI2U4I2S","created_at":"2026-07-30T01:19:41.575114+00:00"},{"alias_kind":"pith_short_8","alias_value":"OMX2R4TN","created_at":"2026-07-30T01:19:41.575114+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/OMX2R4TNZI2U4I2SRFOVNQUDL4","json":"https://pith.science/pith/OMX2R4TNZI2U4I2SRFOVNQUDL4.json","graph_json":"https://pith.science/api/pith-number/OMX2R4TNZI2U4I2SRFOVNQUDL4/graph.json","events_json":"https://pith.science/api/pith-number/OMX2R4TNZI2U4I2SRFOVNQUDL4/events.json","paper":"https://pith.science/paper/OMX2R4TN"},"agent_actions":{"view_html":"https://pith.science/pith/OMX2R4TNZI2U4I2SRFOVNQUDL4","download_json":"https://pith.science/pith/OMX2R4TNZI2U4I2SRFOVNQUDL4.json","view_paper":"https://pith.science/paper/OMX2R4TN","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2603.15684&json=true","fetch_graph":"https://pith.science/api/pith-number/OMX2R4TNZI2U4I2SRFOVNQUDL4/graph.json","fetch_events":"https://pith.science/api/pith-number/OMX2R4TNZI2U4I2SRFOVNQUDL4/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/OMX2R4TNZI2U4I2SRFOVNQUDL4/action/timestamp_anchor","attest_storage":"https://pith.science/pith/OMX2R4TNZI2U4I2SRFOVNQUDL4/action/storage_attestation","attest_author":"https://pith.science/pith/OMX2R4TNZI2U4I2SRFOVNQUDL4/action/author_attestation","sign_citation":"https://pith.science/pith/OMX2R4TNZI2U4I2SRFOVNQUDL4/action/citation_signature","submit_replication":"https://pith.science/pith/OMX2R4TNZI2U4I2SRFOVNQUDL4/action/replication_record"}},"created_at":"2026-07-30T01:19:41.575114+00:00","updated_at":"2026-07-30T01:19:41.575114+00:00"}