{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:EQVHYHBPQTFKTER2PPBYSXMIVQ","short_pith_number":"pith:EQVHYHBP","schema_version":"1.0","canonical_sha256":"242a7c1c2f84caa9923a7bc3895d88ac1b81c77d5c22d2f21ed7baba6654b94b","source":{"kind":"arxiv","id":"2503.10619","version":5},"attestation_state":"computed","paper":{"title":"Tempest: Autonomous Multi-Turn Jailbreaking of Large Language Models with Tree Search","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.CR"],"primary_cat":"cs.AI","authors_text":"Andy Zhou, Ron Arel","submitted_at":"2025-03-13T17:57:32Z","abstract_excerpt":"We introduce Tempest, a multi-turn adversarial framework that models the gradual erosion of Large Language Model (LLM) safety through a tree search perspective. Unlike single-turn jailbreaks that rely on one meticulously engineered prompt, Tempest expands the conversation at each turn in a breadth-first fashion, branching out multiple adversarial prompts that exploit partial compliance from previous responses. By tracking these incremental policy leaks and re-injecting them into subsequent queries, Tempest reveals how minor concessions can accumulate into fully disallowed outputs. Evaluations "},"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":"2503.10619","kind":"arxiv","version":5},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2025-03-13T17:57:32Z","cross_cats_sorted":["cs.CL","cs.CR"],"title_canon_sha256":"71a358e549e7b2f2347ede9857dad70c1a7691edcf1903365a7003aa995f34a9","abstract_canon_sha256":"d6176797e923fb322969238ebedc3b376e00f3369ef246edf7246896cf28400d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:10:54.276014Z","signature_b64":"hhaeb17KCLI1OqC+hsCKmJ8aq/bmXF4wEbD27x7xswh2rfJvm4yeN9Sc5B4OL7r2XLua/cHmZFeT4ez30QahAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"242a7c1c2f84caa9923a7bc3895d88ac1b81c77d5c22d2f21ed7baba6654b94b","last_reissued_at":"2026-07-05T11:10:54.275491Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:10:54.275491Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Tempest: Autonomous Multi-Turn Jailbreaking of Large Language Models with Tree Search","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.CR"],"primary_cat":"cs.AI","authors_text":"Andy Zhou, Ron Arel","submitted_at":"2025-03-13T17:57:32Z","abstract_excerpt":"We introduce Tempest, a multi-turn adversarial framework that models the gradual erosion of Large Language Model (LLM) safety through a tree search perspective. Unlike single-turn jailbreaks that rely on one meticulously engineered prompt, Tempest expands the conversation at each turn in a breadth-first fashion, branching out multiple adversarial prompts that exploit partial compliance from previous responses. By tracking these incremental policy leaks and re-injecting them into subsequent queries, Tempest reveals how minor concessions can accumulate into fully disallowed outputs. Evaluations "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.10619","kind":"arxiv","version":5},"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/2503.10619/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":"2503.10619","created_at":"2026-07-05T11:10:54.275559+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.10619v5","created_at":"2026-07-05T11:10:54.275559+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.10619","created_at":"2026-07-05T11:10:54.275559+00:00"},{"alias_kind":"pith_short_12","alias_value":"EQVHYHBPQTFK","created_at":"2026-07-05T11:10:54.275559+00:00"},{"alias_kind":"pith_short_16","alias_value":"EQVHYHBPQTFKTER2","created_at":"2026-07-05T11:10:54.275559+00:00"},{"alias_kind":"pith_short_8","alias_value":"EQVHYHBP","created_at":"2026-07-05T11:10:54.275559+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.20408","citing_title":"NRT-Bench: Benchmarking Multi-Turn Red-Teaming of LLM Operator Agents in Safety-Critical Control Rooms","ref_index":34,"is_internal_anchor":false},{"citing_arxiv_id":"2604.04060","citing_title":"CoopGuard: Stateful Cooperative Agents Safeguarding LLMs Against Evolving Multi-Round Attacks","ref_index":39,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05058","citing_title":"SoK: Robustness in Large Language Models against Jailbreak Attacks","ref_index":102,"is_internal_anchor":false},{"citing_arxiv_id":"2604.10326","citing_title":"Jailbreaking the Matrix: Nullspace Steering for Controlled Model Subversion","ref_index":32,"is_internal_anchor":false},{"citing_arxiv_id":"2604.11309","citing_title":"The Salami Slicing Threat: Exploiting Cumulative Risks in LLM Systems","ref_index":23,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/EQVHYHBPQTFKTER2PPBYSXMIVQ","json":"https://pith.science/pith/EQVHYHBPQTFKTER2PPBYSXMIVQ.json","graph_json":"https://pith.science/api/pith-number/EQVHYHBPQTFKTER2PPBYSXMIVQ/graph.json","events_json":"https://pith.science/api/pith-number/EQVHYHBPQTFKTER2PPBYSXMIVQ/events.json","paper":"https://pith.science/paper/EQVHYHBP"},"agent_actions":{"view_html":"https://pith.science/pith/EQVHYHBPQTFKTER2PPBYSXMIVQ","download_json":"https://pith.science/pith/EQVHYHBPQTFKTER2PPBYSXMIVQ.json","view_paper":"https://pith.science/paper/EQVHYHBP","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.10619&json=true","fetch_graph":"https://pith.science/api/pith-number/EQVHYHBPQTFKTER2PPBYSXMIVQ/graph.json","fetch_events":"https://pith.science/api/pith-number/EQVHYHBPQTFKTER2PPBYSXMIVQ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/EQVHYHBPQTFKTER2PPBYSXMIVQ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/EQVHYHBPQTFKTER2PPBYSXMIVQ/action/storage_attestation","attest_author":"https://pith.science/pith/EQVHYHBPQTFKTER2PPBYSXMIVQ/action/author_attestation","sign_citation":"https://pith.science/pith/EQVHYHBPQTFKTER2PPBYSXMIVQ/action/citation_signature","submit_replication":"https://pith.science/pith/EQVHYHBPQTFKTER2PPBYSXMIVQ/action/replication_record"}},"created_at":"2026-07-05T11:10:54.275559+00:00","updated_at":"2026-07-05T11:10:54.275559+00:00"}