{"paper":{"title":"Can You Trust the Vectors in Your Vector Database? Black-Hole Attack from Embedding Space Defects","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"A few vectors placed near the center of an embedding space can appear in the top results for nearly every query.","cross_cats":["cs.DB"],"primary_cat":"cs.CR","authors_text":"Hanxi Li, Jiale Lao, Jianan Zhou, Junfen Wang, Mingjie Tang, Yang Cao, Yibo Wang, Zhengmao Ye","submitted_at":"2026-04-07T06:21:41Z","abstract_excerpt":"Vector databases serve as the retrieval backbone of modern AI applications, yet their security remains largely unexplored. We propose the Black-Hole Attack, a poisoning attack that injects a small number of malicious vectors near the geometric center of the stored vectors. These injected vectors attract queries like a black hole and frequently appear in the top-k retrieval results for most queries. This attack is enabled by a phenomenon we term centrality-driven hubness: in high-dimensional embedding spaces, vectors near the centroid become nearest neighbors of a disproportionately large numbe"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"The attack shows that vectors in a vector database cannot be blindly trusted: geometric defects in high-dimensional embeddings make retrieval inherently vulnerable.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That high-dimensional embedding spaces in practice have a nearly empty centroid region where vectors exhibit centrality-driven hubness and become nearest neighbors to a disproportionately large number of other vectors.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Injecting a few malicious vectors near the centroid exploits centrality-driven hubness in high-dimensional embeddings, causing them to dominate top-k retrievals in up to 99.85% of cases.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A few vectors placed near the center of an embedding space can appear in the top results for nearly every query.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"340bab1b91fa54ed0907cad9dc5be03dea3fc271ee27860589cf41eca85d5360"},"source":{"id":"2604.05480","kind":"arxiv","version":2},"verdict":{"id":"73851eac-de30-472d-a9c3-ecd9f2879151","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-10T19:35:12.492863Z","strongest_claim":"The attack shows that vectors in a vector database cannot be blindly trusted: geometric defects in high-dimensional embeddings make retrieval inherently vulnerable.","one_line_summary":"Injecting a few malicious vectors near the centroid exploits centrality-driven hubness in high-dimensional embeddings, causing them to dominate top-k retrievals in up to 99.85% of cases.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That high-dimensional embedding spaces in practice have a nearly empty centroid region where vectors exhibit centrality-driven hubness and become nearest neighbors to a disproportionately large number of other vectors.","pith_extraction_headline":"A few vectors placed near the center of an embedding space can appear in the top results for nearly every query."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.05480/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":2,"snapshot_sha256":"2b8d652bc255afd8571e9dcd246bc365e6006f5a5b408c3cea35aaa8018220ac"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}