{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:PMTSRNFZHY7ZY6EF2G4SECS2LF","short_pith_number":"pith:PMTSRNFZ","schema_version":"1.0","canonical_sha256":"7b2728b4b93e3f9c7885d1b9220a5a594fbc854525d5f79450e9b21698dfed56","source":{"kind":"arxiv","id":"2505.17147","version":1},"attestation_state":"computed","paper":{"title":"MTSA: Multi-turn Safety Alignment for LLMs through Multi-round Red-teaming","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CR","authors_text":"Daojing He, Jing Li, Jun Yu, Min Zhang, Weiyang Guo, Wenya Wang, Yu Li","submitted_at":"2025-05-22T08:22:57Z","abstract_excerpt":"The proliferation of jailbreak attacks against large language models (LLMs) highlights the need for robust security measures. However, in multi-round dialogues, malicious intentions may be hidden in interactions, leading LLMs to be more prone to produce harmful responses. In this paper, we propose the \\textbf{M}ulti-\\textbf{T}urn \\textbf{S}afety \\textbf{A}lignment (\\ourapproach) framework, to address the challenge of securing LLMs in multi-round interactions. It consists of two stages: In the thought-guided attack learning stage, the red-team model learns about thought-guided multi-round jailb"},"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":"2505.17147","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2025-05-22T08:22:57Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"6c2c9198ac2af62bfea04fbff76c5bdcf923973e3bcea206782edef89ba109d9","abstract_canon_sha256":"4b1522d48a4c019bfb4aff6a2ad7b6602bcb7d3503c7d32fdac1bc23daceb47b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:07:47.831273Z","signature_b64":"Iug/tF0tnCi9Ky2bPxb2Knth/gP224sEswXJvKh30k1HF6xedC9znTc43W+0BREDSvycMUY8mCVOYBIOabvPDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"7b2728b4b93e3f9c7885d1b9220a5a594fbc854525d5f79450e9b21698dfed56","last_reissued_at":"2026-07-05T11:07:47.830794Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:07:47.830794Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"MTSA: Multi-turn Safety Alignment for LLMs through Multi-round Red-teaming","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CR","authors_text":"Daojing He, Jing Li, Jun Yu, Min Zhang, Weiyang Guo, Wenya Wang, Yu Li","submitted_at":"2025-05-22T08:22:57Z","abstract_excerpt":"The proliferation of jailbreak attacks against large language models (LLMs) highlights the need for robust security measures. However, in multi-round dialogues, malicious intentions may be hidden in interactions, leading LLMs to be more prone to produce harmful responses. In this paper, we propose the \\textbf{M}ulti-\\textbf{T}urn \\textbf{S}afety \\textbf{A}lignment (\\ourapproach) framework, to address the challenge of securing LLMs in multi-round interactions. It consists of two stages: In the thought-guided attack learning stage, the red-team model learns about thought-guided multi-round jailb"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.17147","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/2505.17147/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":"2505.17147","created_at":"2026-07-05T11:07:47.830855+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.17147v1","created_at":"2026-07-05T11:07:47.830855+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.17147","created_at":"2026-07-05T11:07:47.830855+00:00"},{"alias_kind":"pith_short_12","alias_value":"PMTSRNFZHY7Z","created_at":"2026-07-05T11:07:47.830855+00:00"},{"alias_kind":"pith_short_16","alias_value":"PMTSRNFZHY7ZY6EF","created_at":"2026-07-05T11:07:47.830855+00:00"},{"alias_kind":"pith_short_8","alias_value":"PMTSRNFZ","created_at":"2026-07-05T11:07:47.830855+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.08277","citing_title":"Mitigating Many-shot Jailbreak Attacks with One Single Demonstration","ref_index":19,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/PMTSRNFZHY7ZY6EF2G4SECS2LF","json":"https://pith.science/pith/PMTSRNFZHY7ZY6EF2G4SECS2LF.json","graph_json":"https://pith.science/api/pith-number/PMTSRNFZHY7ZY6EF2G4SECS2LF/graph.json","events_json":"https://pith.science/api/pith-number/PMTSRNFZHY7ZY6EF2G4SECS2LF/events.json","paper":"https://pith.science/paper/PMTSRNFZ"},"agent_actions":{"view_html":"https://pith.science/pith/PMTSRNFZHY7ZY6EF2G4SECS2LF","download_json":"https://pith.science/pith/PMTSRNFZHY7ZY6EF2G4SECS2LF.json","view_paper":"https://pith.science/paper/PMTSRNFZ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.17147&json=true","fetch_graph":"https://pith.science/api/pith-number/PMTSRNFZHY7ZY6EF2G4SECS2LF/graph.json","fetch_events":"https://pith.science/api/pith-number/PMTSRNFZHY7ZY6EF2G4SECS2LF/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/PMTSRNFZHY7ZY6EF2G4SECS2LF/action/timestamp_anchor","attest_storage":"https://pith.science/pith/PMTSRNFZHY7ZY6EF2G4SECS2LF/action/storage_attestation","attest_author":"https://pith.science/pith/PMTSRNFZHY7ZY6EF2G4SECS2LF/action/author_attestation","sign_citation":"https://pith.science/pith/PMTSRNFZHY7ZY6EF2G4SECS2LF/action/citation_signature","submit_replication":"https://pith.science/pith/PMTSRNFZHY7ZY6EF2G4SECS2LF/action/replication_record"}},"created_at":"2026-07-05T11:07:47.830855+00:00","updated_at":"2026-07-05T11:07:47.830855+00:00"}