{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:RPCVHXTSGFIHOWNHCU42GK3TLN","short_pith_number":"pith:RPCVHXTS","schema_version":"1.0","canonical_sha256":"8bc553de7231507759a71539a32b735b5d72f0775ca210e9ebdbbb90ad1e0522","source":{"kind":"arxiv","id":"2412.05734","version":2},"attestation_state":"computed","paper":{"title":"LeakAgent: RL-based Red-teaming Agent for LLM Privacy Leakage","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CR","authors_text":"Dawn Song, Wenbo Guo, Xian Wu, Xuandong Zhao, Ye Yu, Yuzhou Nie, Zhun Wang","submitted_at":"2024-12-07T20:09:01Z","abstract_excerpt":"Recent studies have discovered that large language models (LLM) may be ``fooled'' to output private information, including training data, system prompts, and personally identifiable information, under carefully crafted adversarial prompts. Existing red-teaming approaches for privacy leakage either rely on manual efforts or focus solely on system prompt extraction, making them ineffective for severe risks of training data leakage. We propose LeakAgent, a novel black-box red-teaming framework for LLM privacy leakage. Our framework trains an open-source LLM through reinforcement learning as the a"},"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":"2412.05734","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CR","submitted_at":"2024-12-07T20:09:01Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"d13cddfb591e1f204d00a1fc38281247a4ea08c2523d1d758e0d6948d2d678d9","abstract_canon_sha256":"1d496de25e9657eb09b9fba4436c40fbe2e55d4a6dce7b9618c0f19a08271259"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:50:29.344908Z","signature_b64":"rm04PO9LRR2efXo+IEcPEfH1AmwCSrHejYY4f2YH+Hc4P6NIhjkp6KDXlWMrelfVK8sYkoTHmw8zB6XI8w15DA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"8bc553de7231507759a71539a32b735b5d72f0775ca210e9ebdbbb90ad1e0522","last_reissued_at":"2026-07-05T11:50:29.344419Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:50:29.344419Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"LeakAgent: RL-based Red-teaming Agent for LLM Privacy Leakage","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CR","authors_text":"Dawn Song, Wenbo Guo, Xian Wu, Xuandong Zhao, Ye Yu, Yuzhou Nie, Zhun Wang","submitted_at":"2024-12-07T20:09:01Z","abstract_excerpt":"Recent studies have discovered that large language models (LLM) may be ``fooled'' to output private information, including training data, system prompts, and personally identifiable information, under carefully crafted adversarial prompts. Existing red-teaming approaches for privacy leakage either rely on manual efforts or focus solely on system prompt extraction, making them ineffective for severe risks of training data leakage. We propose LeakAgent, a novel black-box red-teaming framework for LLM privacy leakage. Our framework trains an open-source LLM through reinforcement learning as the a"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2412.05734","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/2412.05734/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":"2412.05734","created_at":"2026-07-05T11:50:29.344478+00:00"},{"alias_kind":"arxiv_version","alias_value":"2412.05734v2","created_at":"2026-07-05T11:50:29.344478+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2412.05734","created_at":"2026-07-05T11:50:29.344478+00:00"},{"alias_kind":"pith_short_12","alias_value":"RPCVHXTSGFIH","created_at":"2026-07-05T11:50:29.344478+00:00"},{"alias_kind":"pith_short_16","alias_value":"RPCVHXTSGFIHOWNH","created_at":"2026-07-05T11:50:29.344478+00:00"},{"alias_kind":"pith_short_8","alias_value":"RPCVHXTS","created_at":"2026-07-05T11:50:29.344478+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.06914","citing_title":"DPAgent-in-the-Middle: Agentic Defense and Repair Against AI-Groomed Deceptive Patterns","ref_index":35,"is_internal_anchor":false},{"citing_arxiv_id":"2606.07539","citing_title":"Prompt Governance? On Governing Technologies Governed by Natural Language","ref_index":246,"is_internal_anchor":false},{"citing_arxiv_id":"2603.09002","citing_title":"Security Considerations for Multi-agent Systems","ref_index":288,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/RPCVHXTSGFIHOWNHCU42GK3TLN","json":"https://pith.science/pith/RPCVHXTSGFIHOWNHCU42GK3TLN.json","graph_json":"https://pith.science/api/pith-number/RPCVHXTSGFIHOWNHCU42GK3TLN/graph.json","events_json":"https://pith.science/api/pith-number/RPCVHXTSGFIHOWNHCU42GK3TLN/events.json","paper":"https://pith.science/paper/RPCVHXTS"},"agent_actions":{"view_html":"https://pith.science/pith/RPCVHXTSGFIHOWNHCU42GK3TLN","download_json":"https://pith.science/pith/RPCVHXTSGFIHOWNHCU42GK3TLN.json","view_paper":"https://pith.science/paper/RPCVHXTS","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2412.05734&json=true","fetch_graph":"https://pith.science/api/pith-number/RPCVHXTSGFIHOWNHCU42GK3TLN/graph.json","fetch_events":"https://pith.science/api/pith-number/RPCVHXTSGFIHOWNHCU42GK3TLN/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/RPCVHXTSGFIHOWNHCU42GK3TLN/action/timestamp_anchor","attest_storage":"https://pith.science/pith/RPCVHXTSGFIHOWNHCU42GK3TLN/action/storage_attestation","attest_author":"https://pith.science/pith/RPCVHXTSGFIHOWNHCU42GK3TLN/action/author_attestation","sign_citation":"https://pith.science/pith/RPCVHXTSGFIHOWNHCU42GK3TLN/action/citation_signature","submit_replication":"https://pith.science/pith/RPCVHXTSGFIHOWNHCU42GK3TLN/action/replication_record"}},"created_at":"2026-07-05T11:50:29.344478+00:00","updated_at":"2026-07-05T11:50:29.344478+00:00"}