[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-researchers-build-a-way-to-search-private-data-without-peeking":10,"sections":35},{"siteName":4,"siteTagline":5,"publisherName":4,"contactEmail":6},"The Revision","Tech news, decoded.","editor@therevision.news",{"gaMeasurementId":8,"adsenseClientId":9},"G-ZW2MV82GYR","ca-pub-8533917693782264",{"article":11},{"id":12,"slug":13,"title":14,"dek":15,"body_md":16,"tags_json":17,"published_at":18,"created_at":19,"updated_at":20,"status":21,"review_note":22,"review_notes":23,"image_url":22,"persona_id":22,"persona_name":22,"section":24,"tags":25,"sources":30,"feedback":34,"feedback_at":22,"cost_usd":34,"total_tokens":34},8469,"researchers-build-a-way-to-search-private-data-without-peeking","Researchers Build a Way to Search Private Data Without Peeking","A new retrieval scheme called PILLAR lets AI systems fetch relevant documents from a server's corpus without revealing what the user searched for.","AI search tools that dig through someone else's documents usually mean that someone else sees exactly what you asked.\n\nResearchers have built PILLAR, a privacy-preserving retrieval-augmented-generation system that lets a client pull the most relevant documents from a server-held corpus without the server learning the query's terms or even which documents got fetched. It works in two stages instead of one. A sparse stage runs a small, fixed number of private-information-retrieval queries against a precomputed index of BM25 relevance scores, narrowing the corpus down to candidates that share terms with the query. A dense stage then retrieves only those candidates' embeddings and reranks them locally, skipping the expensive round-trip queries that earlier private-retrieval systems needed for every step of an approximate nearest-neighbor search. The team built two versions: PILLAR-Bin, a single-round design using hashed posting lists for speed, and PILLAR-Tree, which uses an oblivious tree traversal and cuckoo hashing for better retrieval quality.\n\nThat two-stage split matters because it fixes the actual bottleneck. Earlier private RAG systems leaned entirely on dense retrieval, which meant many rounds of costly PIR queries and a real hit to both speed and result quality. By handling the coarse filtering with lexical search and saving the expensive private queries for a shrunk candidate pool, PILLAR reportedly beats prior state-of-the-art private retrieval schemes on latency, quality, or both, depending on which variant you pick.\n\nThis is the kind of plumbing that matters if RAG ever gets used on genuinely sensitive corpora - legal databases, medical records, internal company wikis - where handing your query to the host is not an option. It is also, for now, a paper on arXiv, not a shipping feature, and the gap between beating benchmarks and working in a production pipeline is exactly where most privacy tech goes to die.","[\"rag\",\"privacy\",\"information-retrieval\",\"ai-research\"]","2026-09-30T04:00:00.000Z","2026-09-30T05:35:59.951Z","2026-09-30T05:36:05.715Z","published",null,[],"ai",[26,27,28,29],"rag","privacy","information-retrieval","ai-research",[31],{"name":32,"url":33},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.36326",0,{"sections":36},[37,40,44,48,53,58,63,68,73,78,83,88,93,98],{"name":38,"slug":24,"count":39,"latest_published_at":18},"AI",5029,{"name":41,"slug":42,"count":43,"latest_published_at":18},"Security","security",780,{"name":45,"slug":46,"count":47,"latest_published_at":18},"Policy","policy",417,{"name":49,"slug":50,"count":51,"latest_published_at":52},"Deals","deals",284,"2026-09-29T21:00:00.000Z",{"name":54,"slug":55,"count":56,"latest_published_at":57},"Hardware","hardware",194,"2026-09-29T13:16:04.000Z",{"name":59,"slug":60,"count":61,"latest_published_at":62},"Science","science",154,"2026-09-28T13:19:18.000Z",{"name":64,"slug":65,"count":66,"latest_published_at":67},"Consumer Tech","consumer-tech",142,"2026-09-29T18:38:03.000Z",{"name":69,"slug":70,"count":71,"latest_published_at":72},"Software","software",91,"2026-09-25T20:55:00.000Z",{"name":74,"slug":75,"count":76,"latest_published_at":77},"Dev Tools","dev-tools",89,"2026-09-29T17:15:00.000Z",{"name":79,"slug":80,"count":81,"latest_published_at":82},"Startups","startups",83,"2026-09-29T21:51:36.000Z",{"name":84,"slug":85,"count":86,"latest_published_at":87},"General","general",49,"2026-09-28T16:44:57.000Z",{"name":89,"slug":90,"count":91,"latest_published_at":92},"Gaming","gaming",48,"2026-09-25T18:35:21.000Z",{"name":94,"slug":95,"count":96,"latest_published_at":97},"Reviews","reviews",31,"2026-09-28T14:31:34.000Z",{"name":99,"slug":100,"count":101,"latest_published_at":102},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]