[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-researchers-build-a-fix-for-ai-agents-that-remember-outdated-facts":10,"sections":34},{"siteName":4,"siteTagline":5,"publisherName":4,"contactEmail":6},"The Revision","Tech news,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":29,"feedback":33,"feedback_at":22,"cost_usd":33,"total_tokens":33},3420,"researchers-build-a-fix-for-ai-agents-that-remember-outdated-facts","Researchers Build a Fix for AI Agents That Remember Outdated Facts","A new memory overlay called ATMA targets \"ghost memory,\" the bug that lets stale and current facts coexist in an AI agent's recall system.","AI agents that track long-running conversations can quietly answer questions with facts that stopped being true months ago.\n\nResearchers introduced ATMA, a state-aware overlay designed to sit on top of existing long-term memory systems rather than replace them. The core problem they document is what they call ghost memory: old facts, current facts, and in-transition facts pile up together in a memory bank, and when the agent goes to retrieve something, it can pull from any layer without knowing which is authoritative. ATMA addresses this by keeping superseded and transitional records in place but tagging them, then assembling labeled evidence packets before handing context to the answer model. The team tested the approach on Graphiti, an existing memory system, and saw conflict accuracy improve by 0.240 absolute on their new benchmark, LTP, and temporal F1 jump from 0.0295 to 0.1705 on the LoCoMo dataset.\n\nThe deeper contribution may be methodological. The researchers argue that measuring only final question-answering accuracy masks where memory failures actually occur - a system can appear to perform well overall while routinely botching time-sensitive facts. They propose evaluating bank maintenance, retrieval, and answer-time resolution as three separate failure modes, which would give developers a clearer picture of what to fix.\n\nLong-term memory has become a quiet battleground for agent frameworks, with every major AI lab and several open-source projects racing to make assistants that persist usefully across sessions. Most benchmarks still reward getting the right answer without asking how close the system came to getting it wrong for the wrong reasons - a gap this work tries to close, even if the gains so far are host-dependent and not yet universal.","[\"ai\",\"agents\",\"memory\",\"benchmarks\"]","2026-07-03T04:00:00.000Z","2026-07-03T05:38:34.581Z","2026-07-03T05:38:37.483Z","published",null,[],"ai",[24,26,27,28],"agents","memory","benchmarks",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.01935",0,{"sections":35},[36,40,45,50,55,60,65,70,75,80,85,89,94,99],{"name":37,"slug":24,"count":38,"latest_published_at":39},"AI",2590,"2026-07-16T04:00:00.000Z",{"name":41,"slug":42,"count":43,"latest_published_at":44},"Security","security",294,"2026-07-15T19:59:48.000Z",{"name":46,"slug":47,"count":48,"latest_published_at":49},"Deals","deals",179,"2026-06-29T20:02:07.000Z",{"name":51,"slug":52,"count":53,"latest_published_at":54},"Policy","policy",158,"2026-07-16T00:02:48.000Z",{"name":56,"slug":57,"count":58,"latest_published_at":59},"Hardware","hardware",122,"2026-07-14T19:46:26.000Z",{"name":61,"slug":62,"count":63,"latest_published_at":64},"Consumer Tech","consumer-tech",93,"2026-07-13T13:20:48.000Z",{"name":66,"slug":67,"count":68,"latest_published_at":69},"Software","software",70,"2026-07-13T19:52:25.000Z",{"name":71,"slug":72,"count":73,"latest_published_at":74},"Science","science",66,"2026-07-10T10:29:37.000Z",{"name":76,"slug":77,"count":78,"latest_published_at":79},"Dev Tools","dev-tools",59,"2026-07-07T04:00:00.000Z",{"name":81,"slug":82,"count":83,"latest_published_at":84},"Gaming","gaming",41,"2026-07-09T04:00:00.000Z",{"name":86,"slug":87,"count":83,"latest_published_at":88},"Startups","startups","2026-06-29T20:55:50.000Z",{"name":90,"slug":91,"count":92,"latest_published_at":93},"General","general",29,"2026-07-10T22:28:58.000Z",{"name":95,"slug":96,"count":97,"latest_published_at":98},"Reviews","reviews",20,"2026-06-24T12:00:01.000Z",{"name":100,"slug":101,"count":102,"latest_published_at":103},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]