{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:CR7ORBJZJRXDJHPTBUREUWDWDR","short_pith_number":"pith:CR7ORBJZ","schema_version":"1.0","canonical_sha256":"147ee885394c6e349df30d224a58761c482b3144a685beeabcab6831346ab3fb","source":{"kind":"arxiv","id":"2503.06254","version":2},"attestation_state":"computed","paper":{"title":"Poisoned-MRAG: Knowledge Poisoning Attacks to Multimodal Retrieval Augmented Generation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CR","authors_text":"Guiyao Tie, Jiawen Shi, Lichao Sun, Neil Zhenqiang Gong, Pan Zhou, Yinuo Liu, Zenghui Yuan","submitted_at":"2025-03-08T15:46:38Z","abstract_excerpt":"Multimodal retrieval-augmented generation (RAG) enhances the visual reasoning capability of vision-language models (VLMs) by dynamically accessing information from external knowledge bases. In this work, we introduce \\textit{Poisoned-MRAG}, the first knowledge poisoning attack on multimodal RAG systems. Poisoned-MRAG injects a few carefully crafted image-text pairs into the multimodal knowledge database, manipulating VLMs to generate the attacker-desired response to a target query. Specifically, we formalize the attack as an optimization problem and propose two cross-modal attack strategies, d"},"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":"2503.06254","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2025-03-08T15:46:38Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"3ba195702782d399f2587d74bac35813f521e048eff61c61ae59ea2f99a82096","abstract_canon_sha256":"a97ed8e0280ca93d4e815ad5d010177f24528f0be3a6870a0c3e371839523430"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:31:02.060588Z","signature_b64":"eDhK3G2pid69+tdFwRO2+8gG8/N90LXzTuELLBNVuGtivJq0XRJIp/qOqPvhUodutOj9xC6k+DNH7COTBqjNDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"147ee885394c6e349df30d224a58761c482b3144a685beeabcab6831346ab3fb","last_reissued_at":"2026-07-05T10:31:02.060053Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:31:02.060053Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Poisoned-MRAG: Knowledge Poisoning Attacks to Multimodal Retrieval Augmented Generation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CR","authors_text":"Guiyao Tie, Jiawen Shi, Lichao Sun, Neil Zhenqiang Gong, Pan Zhou, Yinuo Liu, Zenghui Yuan","submitted_at":"2025-03-08T15:46:38Z","abstract_excerpt":"Multimodal retrieval-augmented generation (RAG) enhances the visual reasoning capability of vision-language models (VLMs) by dynamically accessing information from external knowledge bases. In this work, we introduce \\textit{Poisoned-MRAG}, the first knowledge poisoning attack on multimodal RAG systems. Poisoned-MRAG injects a few carefully crafted image-text pairs into the multimodal knowledge database, manipulating VLMs to generate the attacker-desired response to a target query. Specifically, we formalize the attack as an optimization problem and propose two cross-modal attack strategies, d"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.06254","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/2503.06254/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":"2503.06254","created_at":"2026-07-05T10:31:02.060115+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.06254v2","created_at":"2026-07-05T10:31:02.060115+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.06254","created_at":"2026-07-05T10:31:02.060115+00:00"},{"alias_kind":"pith_short_12","alias_value":"CR7ORBJZJRXD","created_at":"2026-07-05T10:31:02.060115+00:00"},{"alias_kind":"pith_short_16","alias_value":"CR7ORBJZJRXDJHPT","created_at":"2026-07-05T10:31:02.060115+00:00"},{"alias_kind":"pith_short_8","alias_value":"CR7ORBJZ","created_at":"2026-07-05T10:31:02.060115+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.30365","citing_title":"Mental Damage: Caption Poisoning Attacks on Retrieval-Augmented Text-to-Music Generation","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10253","citing_title":"Knowledge Poisoning Attacks on Medical Multi-Modal Retrieval-Augmented Generation","ref_index":68,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/CR7ORBJZJRXDJHPTBUREUWDWDR","json":"https://pith.science/pith/CR7ORBJZJRXDJHPTBUREUWDWDR.json","graph_json":"https://pith.science/api/pith-number/CR7ORBJZJRXDJHPTBUREUWDWDR/graph.json","events_json":"https://pith.science/api/pith-number/CR7ORBJZJRXDJHPTBUREUWDWDR/events.json","paper":"https://pith.science/paper/CR7ORBJZ"},"agent_actions":{"view_html":"https://pith.science/pith/CR7ORBJZJRXDJHPTBUREUWDWDR","download_json":"https://pith.science/pith/CR7ORBJZJRXDJHPTBUREUWDWDR.json","view_paper":"https://pith.science/paper/CR7ORBJZ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.06254&json=true","fetch_graph":"https://pith.science/api/pith-number/CR7ORBJZJRXDJHPTBUREUWDWDR/graph.json","fetch_events":"https://pith.science/api/pith-number/CR7ORBJZJRXDJHPTBUREUWDWDR/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/CR7ORBJZJRXDJHPTBUREUWDWDR/action/timestamp_anchor","attest_storage":"https://pith.science/pith/CR7ORBJZJRXDJHPTBUREUWDWDR/action/storage_attestation","attest_author":"https://pith.science/pith/CR7ORBJZJRXDJHPTBUREUWDWDR/action/author_attestation","sign_citation":"https://pith.science/pith/CR7ORBJZJRXDJHPTBUREUWDWDR/action/citation_signature","submit_replication":"https://pith.science/pith/CR7ORBJZJRXDJHPTBUREUWDWDR/action/replication_record"}},"created_at":"2026-07-05T10:31:02.060115+00:00","updated_at":"2026-07-05T10:31:02.060115+00:00"}