[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-85981-en":3,"doc-seo-85981-105":30,"detail-sidebar-cat-0-en-105":91},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":21,"is_downloadable":21,"audit_status":21,"page_count":22,"language":23,"language_code":24,"site_id":25,"html_lang":24,"table_of_contents":26,"faqs":27,"seo_title":13,"seo_description":14,"update_tm":28,"read_time":29},85981,1374391975076,"Riley","https://ap-avatar.wpscdn.com/avatar/14000253ca4ec9f6853?x-image-process=image/resize,m_fixed,w_180,h_180&k=1783305029341752051",8,"Research & Report","The Compliance Trap Diagnosing How AI Agents Consume Conflicting Memory","Memory is becoming a core component of long-horizon AI agents, enabling reuse of prior experience in interactive environments such as web browsers and software tools. Prior research focuses on a supply pipeline—what memories to write, store, and retrieve—while under-explaining how agents consume retrieved memory across multi-step action trajectories. This work proposes Entry–Propagation–Recovery (E-P-R) to trace where memory first changes actions, whether effects propagate, and if agents recover after divergence. Experiments on WebArena and MemTrapBench show early entry causes adoption of conflicting memories, repeated exposure amplifies error, recovery is weak, and compliance can collapse success rates, with stronger agents suffering larger absolute damage.","arXiv :2607 . 10608v 1 [ cs .AI] 12 Jul 2026  \nThe Compliance Trap: Diagnosing How AI Agents Consume Conflicting Memory  \nYixiong Chen 1 , Xinyi Bai2 , Alan Yuille 1  \n1Johns Hopkins University 2 Cornell University  \n[ychen646@jh.edu](ychen646@jh.edu)  \nAbstract  \nMemory is becoming a core component of long-horizon AI agents, allowing agents to reuse past experience when operating web browsers, software tools, and other interactive environments. Existing work mostly treats memory as a supply problem, asking what experience to write, how to store it, and which entry to retrieve for the next task. Yet we still lack a clear account of how models consume retrieved memory across a multi-step action trajectory. This consumption process matters because it determines not only what memories should be retrieved, but also what models and control policies are needed to use them safely. To diagnose this process, we propose Entry–Propagation–Recovery (E-P-R), a trajectory-level framework that asks where memory first changes an action, whether that change carries forward, and whether the agent can recover after leaving a correct path. We instantiate E-P-R on WebArena and on MemTrapBench, a controlled benchmark we build to isolate these phases. We find that the main failure often begins at entry: agents adopt conflicting memory at the first exposed decision point even when it is task-wrong. Repeated exposure then amplifies this early error, while recovery after divergence is weak. Together, these effects create a compliance trap: across models, conflicting memory induces similar compliance rates, but once agents comply, their success rates collapse to a low floor. Stronger agents therefore suffer larger absolute damage because each compliance event erases more baseline capability. These results suggest that memory-augmented agents should be evaluated not only by retrieval quality or final success rate, but by how they consume memory throughout the trajectory.  \n1 Introduction  \nLong-horizon agents are increasingly used to operate web browsers, file systems, software tools, and operating systems [33, 38] . As these tasks span many observations and actions, agents need a way to reuse past experience instead of solving each task from scratch. Memory has become a common mechanism for this purpose. Trajectories from past episodes can be summarized, stored, and retrieved at the start of a new task to guide later decisions [31, 49, 35, 19, 26, 43, 51] . This makes memory a central component of practical agent systems, especially when the task is too long to fit all useful experience into the model context or too costly to reason through repeatedly.  \nMost existing work studies agent memory from the supply side. Systems such as Reflexion [31], ExpeL [49], AWM [35], WebCoach [19], MemGPT [26], A-MEM [43], MemoryBank [51], and RAP [13] differ in how they write, retrieve, or summarize experience, but they share an implicit assumption: once a memory is placed in context, the agent policy will decide how to use it. However, the agent that consumes memory is still an imperfect language model policy, and retrieval does not determine how the model will use the retrieved content. The useful memory can be ignored, followed too literally, or applied again after the state has changed. As a result, a memory that looks plausible to a retriever can be dangerous if the model adopts it at the wrong point or fails to recover from  \nPreprint.  \nA. Existing focus: memory supply  \nPast trajectories  \nWrite / store  \n| Retrieve |\n| --- |\n|  |\n| Memory in prompt |\n\nPrior work asks what memory reaches the agent.  \nB. Our focus: memory consumption  \nRetrieved memory  \n no-memory baseline  \nStart  Later Outcome  \nmemory-affected trajectory  \nWe ask how retrieved memory changes the agent’s later decisions.  \nFigure 1: Motivation and Overview. (A) Prior work studies the memory supply side. (B) We analyze  \nmemory consumption: how trajectories change with/without memory injectio","cbCaipogONboUgIX","https://ap.wps.com/l/cbCaipogONboUgIX","pdf",447831,4,1,25,"English","en",105,"# Introduction\n## Existing focus: memory supply\n## Our focus: memory consumption\n## Entry–Propagation–Recovery (E-P-R)\n## Experimental evaluation and findings","[{\"question\":\"What problem does the paper address about memory-augmented AI agents?\",\"answer\":\"It addresses how retrieved memory is consumed across multi-step trajectories, not just how memory is retrieved or inserted into prompts.\"},{\"question\":\"How does the Entry–Propagation–Recovery (E-P-R) framework work?\",\"answer\":\"E-P-R decomposes memory consumption into three phases: Entry (whether the first exposed decision changes), Propagation (whether the change persists under continued exposure), and Recovery (whether the agent can return to a correct trajectory after divergence).\"},{\"question\":\"What does the paper find about conflicting memory effects?\",\"answer\":\"The main failures start at entry, where agents adopt conflicting memory at the first exposed decision point; repeated exposure amplifies the early error and recovery after divergence is weak.\"}]",1784207552,63,{"code":4,"msg":31,"data":32},"ok",{"site_id":25,"language":24,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":28},"the-compliance-trap-diagnosing-how-ai-agents-consume-conflicting-memory","",{"@graph":36,"@context":85},[37,53,68],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,48,51],{"item":41,"name":42,"@type":43,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":47},"https://docshare.wps.com/document/","Document",2,{"item":49,"name":12,"@type":43,"position":50},"https://docshare.wps.com/document/research-report/",3,{"item":52,"name":13,"@type":43,"position":20},"https://docshare.wps.com/document/the-compliance-trap-diagnosing-how-ai-agents-consume-conflicting-memory/85981/",{"url":52,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":24,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":41,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-26","2026-07-16",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"What problem does the paper address about memory-augmented AI agents?","Question",{"text":75,"@type":76},"It addresses how retrieved memory is consumed across multi-step trajectories, not just how memory is retrieved or inserted into prompts.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does the Entry–Propagation–Recovery (E-P-R) framework work?",{"text":80,"@type":76},"E-P-R decomposes memory consumption into three phases: Entry (whether the first exposed decision changes), Propagation (whether the change persists under continued exposure), and Recovery (whether the agent can return to a correct trajectory after divergence).",{"name":82,"@type":73,"acceptedAnswer":83},"What does the paper find about conflicting memory effects?",{"text":84,"@type":76},"The main failures start at entry, where agents adopt conflicting memory at the first exposed decision point; repeated exposure amplifies the early error and recovery after divergence is weak.","https://schema.org",{"og:url":52,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":52},"index,follow",{"doc_id":7,"site_id":25},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,123,128,131,135],{"id":21,"doc_module":4,"doc_module_name":46,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":47,"doc_module":4,"doc_module_name":46,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":20,"doc_module":4,"doc_module_name":46,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":46,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":46,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":46,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":46,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":46,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":46,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":46,"category_name":137,"show_sort_weight":106,"slug":138},19,"General","general"]