{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:EQ4OBNMDOJZBKJT2OYKF67CDLS","short_pith_number":"pith:EQ4OBNMD","schema_version":"1.0","canonical_sha256":"2438e0b583727215267a76145f7c435cb7a6ce80f219c7e26a76117ef0d32f76","source":{"kind":"arxiv","id":"2402.17840","version":3},"attestation_state":"computed","paper":{"title":"Follow My Instruction and Spill the Beans: Scalable Data Extraction from Retrieval-Augmented Generation Systems","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CR","cs.LG"],"primary_cat":"cs.CL","authors_text":"Eric Xing, Hanlin Zhang, Himabindu Lakkaraju, Sham Kakade, Zhenting Qi","submitted_at":"2024-02-27T19:08:05Z","abstract_excerpt":"Retrieval-Augmented Generation (RAG) improves pre-trained models by incorporating external knowledge at test time to enable customized adaptation. We study the risk of datastore leakage in Retrieval-In-Context RAG Language Models (LMs). We show that an adversary can exploit LMs' instruction-following capabilities to easily extract text data verbatim from the datastore of RAG systems built with instruction-tuned LMs via prompt injection. The vulnerability exists for a wide range of modern LMs that span Llama2, Mistral/Mixtral, Vicuna, SOLAR, WizardLM, Qwen1.5, and Platypus2, and the exploitabil"},"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":"2402.17840","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-02-27T19:08:05Z","cross_cats_sorted":["cs.AI","cs.CR","cs.LG"],"title_canon_sha256":"d7bb4b1e3d585c50989daff44fcf6d3d1bcf7a36edc7134569dfc0bf545166a5","abstract_canon_sha256":"0a7d3b7edc4087721353a6ca9ba50e470b45a12e0c56a19f14c3404d3b02708e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:16:13.195236Z","signature_b64":"MuBD9dHCV3ZqPatO7EndCRLrPe0mi9aRSpu3gDDeLJy6qOHcLObmvrnamiKbcIi+xAa5rd8v2q+2A4WVA+DRBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"2438e0b583727215267a76145f7c435cb7a6ce80f219c7e26a76117ef0d32f76","last_reissued_at":"2026-07-05T09:16:13.194741Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:16:13.194741Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Follow My Instruction and Spill the Beans: Scalable Data Extraction from Retrieval-Augmented Generation Systems","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CR","cs.LG"],"primary_cat":"cs.CL","authors_text":"Eric Xing, Hanlin Zhang, Himabindu Lakkaraju, Sham Kakade, Zhenting Qi","submitted_at":"2024-02-27T19:08:05Z","abstract_excerpt":"Retrieval-Augmented Generation (RAG) improves pre-trained models by incorporating external knowledge at test time to enable customized adaptation. We study the risk of datastore leakage in Retrieval-In-Context RAG Language Models (LMs). We show that an adversary can exploit LMs' instruction-following capabilities to easily extract text data verbatim from the datastore of RAG systems built with instruction-tuned LMs via prompt injection. The vulnerability exists for a wide range of modern LMs that span Llama2, Mistral/Mixtral, Vicuna, SOLAR, WizardLM, Qwen1.5, and Platypus2, and the exploitabil"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2402.17840","kind":"arxiv","version":3},"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/2402.17840/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":"2402.17840","created_at":"2026-07-05T09:16:13.194794+00:00"},{"alias_kind":"arxiv_version","alias_value":"2402.17840v3","created_at":"2026-07-05T09:16:13.194794+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2402.17840","created_at":"2026-07-05T09:16:13.194794+00:00"},{"alias_kind":"pith_short_12","alias_value":"EQ4OBNMDOJZB","created_at":"2026-07-05T09:16:13.194794+00:00"},{"alias_kind":"pith_short_16","alias_value":"EQ4OBNMDOJZBKJT2","created_at":"2026-07-05T09:16:13.194794+00:00"},{"alias_kind":"pith_short_8","alias_value":"EQ4OBNMD","created_at":"2026-07-05T09:16:13.194794+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":7,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.26627","citing_title":"Agents That Know Too Much: A Data-Centric Survey of Privacy in LLM Agents","ref_index":94,"is_internal_anchor":false},{"citing_arxiv_id":"2605.27825","citing_title":"MRMMIA: Membership Inference Attacks on Memory in Chat Agents","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2409.10102","citing_title":"Trustworthiness in Retrieval-Augmented Generation Systems: A Survey","ref_index":151,"is_internal_anchor":false},{"citing_arxiv_id":"2502.05206","citing_title":"Safety at Scale: A Comprehensive Survey of Large Model and Agent Safety","ref_index":213,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05818","citing_title":"LeakDojo: Decoding the Leakage Threats of RAG Systems","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2604.09747","citing_title":"ADAM: A Systematic Data Extraction Attack on Agent Memory via Adaptive Querying","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2604.18231","citing_title":"AgenTEE: Confidential LLM Agent Execution on Edge Devices","ref_index":40,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/EQ4OBNMDOJZBKJT2OYKF67CDLS","json":"https://pith.science/pith/EQ4OBNMDOJZBKJT2OYKF67CDLS.json","graph_json":"https://pith.science/api/pith-number/EQ4OBNMDOJZBKJT2OYKF67CDLS/graph.json","events_json":"https://pith.science/api/pith-number/EQ4OBNMDOJZBKJT2OYKF67CDLS/events.json","paper":"https://pith.science/paper/EQ4OBNMD"},"agent_actions":{"view_html":"https://pith.science/pith/EQ4OBNMDOJZBKJT2OYKF67CDLS","download_json":"https://pith.science/pith/EQ4OBNMDOJZBKJT2OYKF67CDLS.json","view_paper":"https://pith.science/paper/EQ4OBNMD","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2402.17840&json=true","fetch_graph":"https://pith.science/api/pith-number/EQ4OBNMDOJZBKJT2OYKF67CDLS/graph.json","fetch_events":"https://pith.science/api/pith-number/EQ4OBNMDOJZBKJT2OYKF67CDLS/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/EQ4OBNMDOJZBKJT2OYKF67CDLS/action/timestamp_anchor","attest_storage":"https://pith.science/pith/EQ4OBNMDOJZBKJT2OYKF67CDLS/action/storage_attestation","attest_author":"https://pith.science/pith/EQ4OBNMDOJZBKJT2OYKF67CDLS/action/author_attestation","sign_citation":"https://pith.science/pith/EQ4OBNMDOJZBKJT2OYKF67CDLS/action/citation_signature","submit_replication":"https://pith.science/pith/EQ4OBNMDOJZBKJT2OYKF67CDLS/action/replication_record"}},"created_at":"2026-07-05T09:16:13.194794+00:00","updated_at":"2026-07-05T09:16:13.194794+00:00"}