[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-84654-en":3,"doc-seo-84654-105":29,"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":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":13,"seo_description":14,"update_tm":27,"read_time":28},84654,3848291630094,"Emma Wilson","https://eur-avatar.wpscdn.com/davatar_085a072bc5b1113ac321206ff7593b45",8,"Research & Report","What LLM Agents Say When No One Is Watching Social Structure and Latent Objective Emergence in Multi Agent Debates","LLM agents increasingly operate in socially structured settings where role, audience, and relational context affect what is advantageous or costly to express. The study tests whether social structure alone—without an explicit prompt objective—changes what an agent publishes versus an off-the-record (OTR) channel under identical conditions. A dual-channel debate framework records public utterances and hidden OTR responses, revealing public–OTR divergence that grows from ~3% to ~40%, consistent across multiple evaluation measures and surveys.","What LLM Agents Say When No One Is Watching: Social Structure and Latent Objective Emergence in Multi-Agent Debates  \nArman Ghaffarizadeh* Independent Researcher  \n[arman@ghaffarizadeh.com](arman@ghaffarizadeh.com)[ ](arman@ghaffarizadeh.com)Aliakbar Izadkhah†  \nIndependent Researcher [izad.aliakbar@gmail.com](izad.aliakbar@gmail.com)  \nDanyal Mohaddes* Independent Researcher [danyal@mohaddes.dev](danyal@mohaddes.dev)  \nShahriar Noroozizadeh†  \nMachine Learning Department & Heinz College Carnegie Mellon University [snoroozi@cs.cmu.edu](snoroozi@cs.cmu.edu)  \narXiv :2607 .02507v 1 [ cs .AI] 2 Jul 2026  \nAbstract  \nLLM agents will increasingly act in socially structured settings where role, audience, and relational context can shape what is advantageous or costly to say. We study whether such social structure, without any explicit objective in the prompt, changes what an agent expresses publicly relative to an off-the-record (OTR) channel elicited under the same condition. We introduce a dual-channel debate framework in which agents produce public utterances that enter the shared history alongside OTR responses that are recorded but never shown to the other participant. Across 10 models, 3 scenarios, and  \n5 variations within each scenario, alignmentinducing settings produce systematic public– OTR divergence in the targeted agent, with its decision divergence rising from a ∼3% baseline to roughly 40% . The effect is consistent across four aggregate analyses: stance, semantic similarity, natural language inference, and survey responses. In some cases the OTR response explicitly attributes public accommodation to relational pressures, such as career risk or sponsorship obligation. The findings suggest that agent evaluation should extend beyond explicit goals and detect emergent objectives. We present a dual-channel evaluation framework and complementary behavioral measures that operationalize this assessment. 1  \n1 Introduction  \nIt is anticipated that large language model (LLM) agents will increasingly be deployed not only as task-completion tools, but as bona fide participants in consequential interactions. We can expect LLM agents to be employed as autonomous representatives of individuals and organizations in diverse circumstances, such as advisory settings, stakeholder  \n* Corresponding authors and equal contribution. †Equal contribution. 1Code and reproducibility details are available at: [https://github.com/danmohad/LLMAgora](https://github.com/danmohad/LLMAgora).  \nnegotiations and institutional deliberations (Abdelnabi et al., 2024 ; Zhou et al., 2024 ; Park et al., 2023) . In such circumstances, agent outputs are communicative acts, addressed to an audience of one or more humans or other LLM agents, and embedded in the particular relational context that exists between the agent and the other parties. When the outcome of an interaction can have reputational, professional, or financial consequences, the relevant question is not simply whether the agent gives a correct response or tool call to a query, or even whether its actions are consistent with its explicitly defined goals, both of which are expected properties of state-of-the-art LLM models (Ouyang et al., 2022 ; Schick et al., 2023 ; Yao et al., 2023 ; Wu et al., 2024, 2025b) . An agent in such a high-stakes representative role would be expected to act with greater nuance, namely, by navigating the implicit aspects of the relational structure within which it is operating, while remaining consistent with the interests it represents.  \nExisting multi-agent debate research largely studies exchanges whose purpose is fixed via external objectives. Agents debate to improve task accuracy, support evaluation, or persuade an audience toward a declared target (Du et al., 2024 ; Liang et al., 2024 ; Estornell and Liu, 2024 ; Chan et al., 2024 ; Zhao et al., 2025 ; Salvi et al., 2025 ; Schoenegger et al., 2025 ; Bozdag et al., 2026a) . This has produced useful evidence about del","cbCail0YWHiOcxwZ","https://ap.wps.com/l/cbCail0YWHiOcxwZ","pdf",4862528,1,68,"English","en",105,"# Abstract\n# Introduction\n## Dual-channel debate framework","[{\"question\":\"What does the paper investigate about social structure in LLM agent communication?\",\"answer\":\"It examines whether social structure, without any explicit objective in the prompt, alters what an agent expresses publicly compared with an off-the-record (OTR) channel.\"},{\"question\":\"How does the study’s dual-channel debate framework work?\",\"answer\":\"Agents generate outputs for a public channel and a separate off-the-record channel; only public utterances are appended to the shared history, while OTR responses are recorded but not shown to the other participant.\"},{\"question\":\"What is the main finding about divergence between public and OTR responses?\",\"answer\":\"In alignment-inducing settings, the targeted agent shows systematic public–OTR divergence, with decision divergence increasing from a baseline around ~3% to roughly ~40%, supported by several aggregate analyses and survey responses.\"}]",1784197507,171,{"code":4,"msg":30,"data":31},"ok",{"site_id":24,"language":23,"slug":32,"title":13,"keywords":33,"description":14,"schema_data":34,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":27},"what-llm-agents-say-when-no-one-is-watching-social-structure-and-latent-objective-emergence-in-multi-agent-debates","",{"@graph":35,"@context":85},[36,53,68],{"@type":37,"itemListElement":38},"BreadcrumbList",[39,43,47,50],{"item":40,"name":41,"@type":42,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":44,"name":45,"@type":42,"position":46},"https://docshare.wps.com/document/","Document",2,{"item":48,"name":12,"@type":42,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":42,"position":52},"https://docshare.wps.com/document/what-llm-agents-say-when-no-one-is-watching-social-structure-and-latent-objective-emergence-in-multi-agent-debates/84654/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":23,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":40,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-17","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 does the paper investigate about social structure in LLM agent communication?","Question",{"text":75,"@type":76},"It examines whether social structure, without any explicit objective in the prompt, alters what an agent expresses publicly compared with an off-the-record (OTR) channel.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does the study’s dual-channel debate framework work?",{"text":80,"@type":76},"Agents generate outputs for a public channel and a separate off-the-record channel; only public utterances are appended to the shared history, while OTR responses are recorded but not shown to the other participant.",{"name":82,"@type":73,"acceptedAnswer":83},"What is the main finding about divergence between public and OTR responses?",{"text":84,"@type":76},"In alignment-inducing settings, the targeted agent shows systematic public–OTR divergence, with decision divergence increasing from a baseline around ~3% to roughly ~40%, supported by several aggregate analyses and survey responses.","https://schema.org",{"og:url":51,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":51},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,123,128,131,135],{"id":20,"doc_module":4,"doc_module_name":45,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":45,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":45,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":45,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":45,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":45,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":45,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":45,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":45,"category_name":137,"show_sort_weight":106,"slug":138},19,"General","general"]