{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:TKTUBDUL7IQHB4JXUFDFJ5PEYL","short_pith_number":"pith:TKTUBDUL","schema_version":"1.0","canonical_sha256":"9aa7408e8bfa2070f137a14654f5e4c2d350c8883f87d0c83207ed5eea069fe4","source":{"kind":"arxiv","id":"2507.18202","version":1},"attestation_state":"computed","paper":{"title":"Safeguarding RAG Pipelines with GMTP: A Gradient-based Masked Token Probability Method for Poisoned Document Detection","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Gary Geunbae Lee, Jonghwi Kim, San Kim, Yejin Jeon","submitted_at":"2025-07-24T08:58:41Z","abstract_excerpt":"Retrieval-Augmented Generation (RAG) enhances Large Language Models (LLMs) by providing external knowledge for accurate and up-to-date responses. However, this reliance on external sources exposes a security risk, attackers can inject poisoned documents into the knowledge base to steer the generation process toward harmful or misleading outputs. In this paper, we propose Gradient-based Masked Token Probability (GMTP), a novel defense method to detect and filter out adversarially crafted documents. Specifically, GMTP identifies high-impact tokens by examining gradients of the retriever's simila"},"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":"2507.18202","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-07-24T08:58:41Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"cb553a97b75de55a697b69e93cedf388b32979c58d2636da7aa2006512382329","abstract_canon_sha256":"acd642dd5fe2d155cb11d1bfb135b94a52e6d2bad3221ee8d96ee385926dcd9b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:42:43.570354Z","signature_b64":"bIsiE8CFP4sFB+sR3H8a72hyLLH6wg4rIVfez1giWBqgih2z9up00loU8ZIcbYNYy3ZEGqc+1RpVhkOpOceQDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9aa7408e8bfa2070f137a14654f5e4c2d350c8883f87d0c83207ed5eea069fe4","last_reissued_at":"2026-07-05T11:42:43.569932Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:42:43.569932Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Safeguarding RAG Pipelines with GMTP: A Gradient-based Masked Token Probability Method for Poisoned Document Detection","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Gary Geunbae Lee, Jonghwi Kim, San Kim, Yejin Jeon","submitted_at":"2025-07-24T08:58:41Z","abstract_excerpt":"Retrieval-Augmented Generation (RAG) enhances Large Language Models (LLMs) by providing external knowledge for accurate and up-to-date responses. However, this reliance on external sources exposes a security risk, attackers can inject poisoned documents into the knowledge base to steer the generation process toward harmful or misleading outputs. In this paper, we propose Gradient-based Masked Token Probability (GMTP), a novel defense method to detect and filter out adversarially crafted documents. Specifically, GMTP identifies high-impact tokens by examining gradients of the retriever's simila"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.18202","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/2507.18202/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":"2507.18202","created_at":"2026-07-05T11:42:43.569990+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.18202v1","created_at":"2026-07-05T11:42:43.569990+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.18202","created_at":"2026-07-05T11:42:43.569990+00:00"},{"alias_kind":"pith_short_12","alias_value":"TKTUBDUL7IQH","created_at":"2026-07-05T11:42:43.569990+00:00"},{"alias_kind":"pith_short_16","alias_value":"TKTUBDUL7IQHB4JX","created_at":"2026-07-05T11:42:43.569990+00:00"},{"alias_kind":"pith_short_8","alias_value":"TKTUBDUL","created_at":"2026-07-05T11:42:43.569990+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/TKTUBDUL7IQHB4JXUFDFJ5PEYL","json":"https://pith.science/pith/TKTUBDUL7IQHB4JXUFDFJ5PEYL.json","graph_json":"https://pith.science/api/pith-number/TKTUBDUL7IQHB4JXUFDFJ5PEYL/graph.json","events_json":"https://pith.science/api/pith-number/TKTUBDUL7IQHB4JXUFDFJ5PEYL/events.json","paper":"https://pith.science/paper/TKTUBDUL"},"agent_actions":{"view_html":"https://pith.science/pith/TKTUBDUL7IQHB4JXUFDFJ5PEYL","download_json":"https://pith.science/pith/TKTUBDUL7IQHB4JXUFDFJ5PEYL.json","view_paper":"https://pith.science/paper/TKTUBDUL","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.18202&json=true","fetch_graph":"https://pith.science/api/pith-number/TKTUBDUL7IQHB4JXUFDFJ5PEYL/graph.json","fetch_events":"https://pith.science/api/pith-number/TKTUBDUL7IQHB4JXUFDFJ5PEYL/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/TKTUBDUL7IQHB4JXUFDFJ5PEYL/action/timestamp_anchor","attest_storage":"https://pith.science/pith/TKTUBDUL7IQHB4JXUFDFJ5PEYL/action/storage_attestation","attest_author":"https://pith.science/pith/TKTUBDUL7IQHB4JXUFDFJ5PEYL/action/author_attestation","sign_citation":"https://pith.science/pith/TKTUBDUL7IQHB4JXUFDFJ5PEYL/action/citation_signature","submit_replication":"https://pith.science/pith/TKTUBDUL7IQHB4JXUFDFJ5PEYL/action/replication_record"}},"created_at":"2026-07-05T11:42:43.569990+00:00","updated_at":"2026-07-05T11:42:43.569990+00:00"}