[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-83274-en":3,"doc-seo-83274-105":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},83274,13056703019662,"Evangeline","https://ap-avatar.wpscdn.com/avatar/be000253a8e92610077?_k=1778726343310543188",8,"Research & Report","User Identity Conditions Moral Wrongness Ratings in Non-Reasoning Large Language Models","This study investigates AI value alignment through a behavioural bottom-up approach, testing whether implicitly conveyed user identity alters large language model moral evaluations. Using a structured multi-turn protocol across 12,000 interactions, two non-reasoning models are prompted without explicit personas or moral stances. The models provide wrongness ratings (0–100) for ten rules from Gert’s moral framework, revealing role-conditioned judgment shifts that reflect the link between profession and rated acts, challenging static moral-bounds assumptions.","arXiv :2607 .07605v 1 [ cs .CY] 8 Jul 2026  \nUser identity conditions moral wrongness ratings in non-reasoning large language models  \nWillem Fourie 1, ∗ Isabel Ray2 Gray Manicom 1  \n1 School for Data Science and Computational Thinking, Stellenbosch University  \n2 Department of Mathematical Sciences, Faculty of Science, Stellenbosch University  \nAbstract  \nThis study adopts a behavioural bottom-up approach to AI value alignment to investigate whether an implicitly conveyed user identity shifts the moral evaluations of large language models (LLMs) . Through a structured, multi-turn conversational protocol across 12,000 interactions, we evaluate AI value alignment in two non-reasoning models, gpt-4.1-mini- 2025-04-14 and gemini-2.5-flash-lite. Rather than instructing the models to adopt a persona or prompting them with explicit moral stances, the user’s professional role is introduced purely through value-neutral reasoning. The models are then asked for wrongness ratings from 0-100 on ten common-morality rules from Gert’s moral framework. The results show that moral judgments vary with the user’s role across both models. While grave-harm acts like killing exhibit a strong ceiling effect, contestable rule-governed acts demonstrate role-conditioned shifts that mirror the relationship between the user’s profession and the act being rated. These findings demonstrate that unintended contextual conditioning via user identity permeates LLM moral evaluations, posing questions for the AI value alignment discourse regarding how to define acceptable bounds for role-based moral divergence. By doing so, the results contribute to reframing the AI value alignment discourse by suggesting future research on dynamic moral bounds rather than static moral principles or rules as frame of reference.  \n1 Introduction  \nThe goal of AI alignment research is to ensure that particularly advanced AI systems operate in line with the intentions and values of their human owners, developers, deployers and users. In their comprehensive AI alignment review, Ji et al. [1] identify four objectives to guide AI alignment research: Robustness, Interpretability, Controllability and Ethicality.  \nOur focus is the ethicality component, framed in the scholarly discourse as AI value alignment [2–8] . At a conceptual level, the ethicality of an AI system means that it does not violate the values of users, developers and society [9–11] .  \nWithin AI value alignment, top-down and bottom-up approaches are typically identified [12] . Top-down approaches focus on defining moral frameworks, rules or principles, such as the golden rule, utilitarianism or Kantian deontology, that the AI system should follow [2, 13] . Top-down approaches have several limitations [14] . Using philosophical moral theories, as in many top-down approaches to AI value alignment, assumes that these abstract theories reflect the moral priorities of individuals [8] . Relatedly, they face challenges when capturing the complexity of individual morality and moral divergence between individuals [15] . Furthermore, these theories are rooted in the Western philosophical tradition, raising questions about their generalisability beyond so-called Western settings.  \n∗ Corresponding author. Email: [willemf@sun.ac.za](willemf@sun.ac.za)  \nBottom-up approaches do not specify moral frameworks, rules or principles. Rather, they focus on ‘the creation of environments or feedback mechanisms that enable agents to learn from human behaviour’ [16] . Bottom-up approaches present challenges of their own. Despite differences between people’s moralities, a fair way to decide on the principles AI systems need to align with is still required [2] .  \nAI value alignment research can also be decomposed methodologically. Using this lens, approaches broadly focusing on the model and its behaviour can be distinguished. AI alignment research focused on the model itself includes a large and growing number of studies using approaches such ","cbCairVoxO0SYZeX","https://ap.wps.com/l/cbCairVoxO0SYZeX","pdf",1402439,1,11,"English","en",105,"# Abstract\n# Introduction\n## AI alignment objectives and ethicality\n## Value alignment approaches (top-down vs bottom-up)\n## Behavioural value alignment and related work\n## Study design and key differences","[{\"question\":\"How does the study test whether user identity affects moral judgments in LLMs?\",\"answer\":\"A structured multi-turn conversational protocol varies an implicitly conveyed professional role, without instructing the models to adopt a persona or explicit moral stances. The models then give wrongness ratings for common-morality rules.\"},{\"question\":\"Which models and rating task are used in the experiments?\",\"answer\":\"The study evaluates two non-reasoning models—gpt-4.1-mini-2025-04-14 and gemini-2.5-flash-lite—asking them to provide wrongness ratings from 0 to 100 on ten rules from Gert’s moral framework.\"},{\"question\":\"What main patterns do the results show across different user roles?\",\"answer\":\"Moral judgments change with the user’s role across both models. Grave-harm acts like killing show a strong ceiling effect, while other rule-governed acts exhibit role-conditioned shifts aligned with the relationship between the user’s profession and the act being rated.\"}]",1784186438,28,{"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},"user-identity-conditions-moral-wrongness-ratings-in-non-reasoning-large-language-models","",{"@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/user-identity-conditions-moral-wrongness-ratings-in-non-reasoning-large-language-models/83274/",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},"How does the study test whether user identity affects moral judgments in LLMs?","Question",{"text":75,"@type":76},"A structured multi-turn conversational protocol varies an implicitly conveyed professional role, without instructing the models to adopt a persona or explicit moral stances. The models then give wrongness ratings for common-morality rules.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"Which models and rating task are used in the experiments?",{"text":80,"@type":76},"The study evaluates two non-reasoning models—gpt-4.1-mini-2025-04-14 and gemini-2.5-flash-lite—asking them to provide wrongness ratings from 0 to 100 on ten rules from Gert’s moral framework.",{"name":82,"@type":73,"acceptedAnswer":83},"What main patterns do the results show across different user roles?",{"text":84,"@type":76},"Moral judgments change with the user’s role across both models. Grave-harm acts like killing show a strong ceiling effect, while other rule-governed acts exhibit role-conditioned shifts aligned with the relationship between the user’s profession and the act being rated.","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"]