{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:QAWPCFZXZFSK7PW2Z5VEQ342RW","short_pith_number":"pith:QAWPCFZX","schema_version":"1.0","canonical_sha256":"802cf11737c964afbedacf6a486f9a8da51b15e9edb12a0f2d094ecdc8fd47e6","source":{"kind":"arxiv","id":"2505.23828","version":1},"attestation_state":"computed","paper":{"title":"Spa-VLM: Stealthy Poisoning Attacks on RAG-based VLM","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CR","authors_text":"Jing Wang, Lei Yu, Minghui Li, Pei Xiaobing, Shengshan Hu, Wei Wan, Yang Wu, Yechao Zhang, Ziqi Zhou","submitted_at":"2025-05-28T07:44:10Z","abstract_excerpt":"With the rapid development of the Vision-Language Model (VLM), significant progress has been made in Visual Question Answering (VQA) tasks. However, existing VLM often generate inaccurate answers due to a lack of up-to-date knowledge. To address this issue, recent research has introduced Retrieval-Augmented Generation (RAG) techniques, commonly used in Large Language Models (LLM), into VLM, incorporating external multi-modal knowledge to enhance the accuracy and practicality of VLM systems. Nevertheless, the RAG in LLM may be susceptible to data poisoning attacks. RAG-based VLM may also face t"},"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":"2505.23828","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.CR","submitted_at":"2025-05-28T07:44:10Z","cross_cats_sorted":[],"title_canon_sha256":"c62fe6a2596e79301158e913ffe173806c5d951a241360283d29a9e2aae4fbd0","abstract_canon_sha256":"815c761e98ea280762dbfa3651ac876c5adf620369642b5a4fedae5a5644d184"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:12:36.059380Z","signature_b64":"eT1uXpfJj9Nvtp2an1f74WiWcxCYZ0CWZ6pe2492bNX19h8KaUH35vSpixcwzrVpDPSl+zy3u5pUOeVMksraCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"802cf11737c964afbedacf6a486f9a8da51b15e9edb12a0f2d094ecdc8fd47e6","last_reissued_at":"2026-07-05T11:12:36.058772Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:12:36.058772Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Spa-VLM: Stealthy Poisoning Attacks on RAG-based VLM","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CR","authors_text":"Jing Wang, Lei Yu, Minghui Li, Pei Xiaobing, Shengshan Hu, Wei Wan, Yang Wu, Yechao Zhang, Ziqi Zhou","submitted_at":"2025-05-28T07:44:10Z","abstract_excerpt":"With the rapid development of the Vision-Language Model (VLM), significant progress has been made in Visual Question Answering (VQA) tasks. However, existing VLM often generate inaccurate answers due to a lack of up-to-date knowledge. To address this issue, recent research has introduced Retrieval-Augmented Generation (RAG) techniques, commonly used in Large Language Models (LLM), into VLM, incorporating external multi-modal knowledge to enhance the accuracy and practicality of VLM systems. Nevertheless, the RAG in LLM may be susceptible to data poisoning attacks. RAG-based VLM may also face t"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.23828","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/2505.23828/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":"2505.23828","created_at":"2026-07-05T11:12:36.058852+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.23828v1","created_at":"2026-07-05T11:12:36.058852+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.23828","created_at":"2026-07-05T11:12:36.058852+00:00"},{"alias_kind":"pith_short_12","alias_value":"QAWPCFZXZFSK","created_at":"2026-07-05T11:12:36.058852+00:00"},{"alias_kind":"pith_short_16","alias_value":"QAWPCFZXZFSK7PW2","created_at":"2026-07-05T11:12:36.058852+00:00"},{"alias_kind":"pith_short_8","alias_value":"QAWPCFZX","created_at":"2026-07-05T11:12:36.058852+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.19736","citing_title":"VFACamou: View-Fused Adversarial Camouflage for Environment-Adaptive Physical Evasion","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10253","citing_title":"Knowledge Poisoning Attacks on Medical Multi-Modal Retrieval-Augmented Generation","ref_index":84,"is_internal_anchor":false},{"citing_arxiv_id":"2604.23105","citing_title":"Transferable Physical-World Adversarial Patches Against Object Detection in Autonomous Driving","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2604.22552","citing_title":"Transferable Physical-World Adversarial Patches Against Pedestrian Detection Models","ref_index":12,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/QAWPCFZXZFSK7PW2Z5VEQ342RW","json":"https://pith.science/pith/QAWPCFZXZFSK7PW2Z5VEQ342RW.json","graph_json":"https://pith.science/api/pith-number/QAWPCFZXZFSK7PW2Z5VEQ342RW/graph.json","events_json":"https://pith.science/api/pith-number/QAWPCFZXZFSK7PW2Z5VEQ342RW/events.json","paper":"https://pith.science/paper/QAWPCFZX"},"agent_actions":{"view_html":"https://pith.science/pith/QAWPCFZXZFSK7PW2Z5VEQ342RW","download_json":"https://pith.science/pith/QAWPCFZXZFSK7PW2Z5VEQ342RW.json","view_paper":"https://pith.science/paper/QAWPCFZX","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.23828&json=true","fetch_graph":"https://pith.science/api/pith-number/QAWPCFZXZFSK7PW2Z5VEQ342RW/graph.json","fetch_events":"https://pith.science/api/pith-number/QAWPCFZXZFSK7PW2Z5VEQ342RW/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/QAWPCFZXZFSK7PW2Z5VEQ342RW/action/timestamp_anchor","attest_storage":"https://pith.science/pith/QAWPCFZXZFSK7PW2Z5VEQ342RW/action/storage_attestation","attest_author":"https://pith.science/pith/QAWPCFZXZFSK7PW2Z5VEQ342RW/action/author_attestation","sign_citation":"https://pith.science/pith/QAWPCFZXZFSK7PW2Z5VEQ342RW/action/citation_signature","submit_replication":"https://pith.science/pith/QAWPCFZXZFSK7PW2Z5VEQ342RW/action/replication_record"}},"created_at":"2026-07-05T11:12:36.058852+00:00","updated_at":"2026-07-05T11:12:36.058852+00:00"}