[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-study-finds-most-ai-agent-defenses-are-free-one-is-not":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},7441,"study-finds-most-ai-agent-defenses-are-free-one-is-not","Study Finds Most AI Agent Defenses Are Free, One Is Not","A new benchmark on benign traffic found three AI agent memory defenses cost nothing measurable, but a reranker hurts accuracy and mislabels real memories.","A new benchmark shows that most defenses against AI agent memory poisoning are free, but the most common one is not.\n\nResearchers built a controlled measurement setup that holds the memory backend, retrieval process, and judge constant while swapping in four different defenses, then ran everything on entirely benign conversations with no attacks in them. Three write-time defenses, input sanitization, provenance checking, and LLM-based anomaly detection, showed no accuracy cost the researchers could distinguish from ordinary pipeline noise. The fourth, a read-time reranker, lowered core accuracy by 4.4 points and wrongly quarantined legitimate memories on 33.6% of adjudicated items, reaching 106 false quarantines in a single conversation, while adding 2.7% token overhead. Stacking all four defenses together did not make things worse. The combined setup's accuracy loss was smaller than the reranker's alone, suggesting the other defenses partly offset what it discarded.\n\nMost security research on these defenses is scored only on how well it stops attacks, so nobody has measured what it costs on the traffic that is not an attack, which in practice is nearly all of it. This paper's finding, that where a defense sits in the pipeline matters more than whether it uses an LLM, is useful for anyone shipping agents with persistent memory. Reranking filters what an agent can see, and every false positive there is a memory the agent silently loses.\n\nThe authors flag their own confidence intervals as borderline, so this reads as an early warning on reranking, not a final verdict.","[\"ai-agents\",\"llm-security\",\"memory-poisoning\",\"ai-research\"]","2026-09-23T04:00:00.000Z","2026-09-23T13:36:48.582Z","2026-09-23T13:36:53.740Z","published",null,[],"ai",[26,27,28,29],"ai-agents","llm-security","memory-poisoning","ai-research",[31],{"name":32,"url":33},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.22818",0,{"sections":36},[37,41,45,50,55,60,65,70,75,80,85,90,95,100],{"name":38,"slug":24,"count":39,"latest_published_at":40},"AI",4347,"2026-09-23T12:00:00.000Z",{"name":42,"slug":43,"count":44,"latest_published_at":18},"Security","security",713,{"name":46,"slug":47,"count":48,"latest_published_at":49},"Policy","policy",371,"2026-09-23T12:00:43.000Z",{"name":51,"slug":52,"count":53,"latest_published_at":54},"Deals","deals",212,"2026-09-23T13:00:46.000Z",{"name":56,"slug":57,"count":58,"latest_published_at":59},"Hardware","hardware",170,"2026-09-23T11:59:23.000Z",{"name":61,"slug":62,"count":63,"latest_published_at":64},"Science","science",134,"2026-09-23T09:00:00.000Z",{"name":66,"slug":67,"count":68,"latest_published_at":69},"Consumer Tech","consumer-tech",110,"2026-09-22T20:00:00.000Z",{"name":71,"slug":72,"count":73,"latest_published_at":74},"Software","software",81,"2026-09-23T09:56:13.000Z",{"name":76,"slug":77,"count":78,"latest_published_at":79},"Dev Tools","dev-tools",79,"2026-09-22T22:21:13.000Z",{"name":81,"slug":82,"count":83,"latest_published_at":84},"Startups","startups",65,"2026-09-22T22:06:48.000Z",{"name":86,"slug":87,"count":88,"latest_published_at":89},"Gaming","gaming",45,"2026-09-22T15:35:06.000Z",{"name":91,"slug":92,"count":93,"latest_published_at":94},"General","general",43,"2026-09-21T23:48:56.000Z",{"name":96,"slug":97,"count":98,"latest_published_at":99},"Reviews","reviews",27,"2026-09-22T13:00:00.000Z",{"name":101,"slug":102,"count":103,"latest_published_at":104},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]