[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-study-finds-proactive-ai-agents-erode-trust-even-when-right":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},8511,"study-finds-proactive-ai-agents-erode-trust-even-when-right","Study Finds Proactive AI Agents Erode Trust Even When Right","A new evaluation testbed and human study show unsolicited AI help can tank user trust immediately, even when the work itself is correct.","Researchers built a training gym for AI agents that get to work before anyone asks them to.\n\nA new paper proposes a framework called 3T for designing proactive AI agents: task capability (can it guess what you need and do it correctly), temporal allocation (does it spend compute wisely, now versus later), and trust (does the user still believe in it after it acts alone). The authors lay out a five-part design space covering task scope, anticipation horizon, activation trigger, processing timing, and intervention depth, then test it with PROACTIVITY-GYM, a simulated testbed running multi-day scenarios with stateful environments and AI-generated personas standing in for real users. Across 23 different model and harness combinations, the team found large performance gaps, and found that AI judges scoring these agents routinely confuse \"did the task well\" with \"did it in a way people trusted.\" A follow-up study with 30 human participants found trust drops sharply the moment an agent's timing or judgment feels off, even when its output is technically correct.\n\nThat last finding matters more than the benchmark itself. Every company building an AI copilot that drafts emails, preps documents, or nudges you mid-task is betting that proactivity reads as helpful rather than presumptuous. This research suggests that bet is fragile: participants forgave being ignored more easily than being interrupted incorrectly, and preferred agents that quietly finish work in the background (what the paper calls sleep-time assistance) over ones that jump in live, even when that background work was imperfect.\n\nIt is the same lesson smartphone notification design learned a decade ago: being right is not the same as being wanted, and an agent that guesses your intent wrong at the wrong moment can cost more trust than it earns in usefulness.","[\"proactive ai\",\"ai agents\",\"user trust\",\"ai research\"]","2026-09-30T04:00:00.000Z","2026-09-30T08:15:32.621Z","2026-09-30T08:15:37.337Z","published",null,[],"ai",[26,27,28,29],"proactive ai","ai agents","user trust","ai research",[31],{"name":32,"url":33},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.37267",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",5028,{"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"]