[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-0-en-105":3,"doc-seo-327634-105":59,"doc-detail-327634-en":131},{"code":4,"msg":5,"data":6},0,"success",[7,13,18,23,28,33,38,43,48,51,55],{"id":8,"doc_module":4,"doc_module_name":9,"category_name":10,"show_sort_weight":11,"slug":12},1,"Document","Story & Novel",90,"story-novel",{"id":14,"doc_module":4,"doc_module_name":9,"category_name":15,"show_sort_weight":16,"slug":17},2,"Literature",80,"literature",{"id":19,"doc_module":4,"doc_module_name":9,"category_name":20,"show_sort_weight":21,"slug":22},4,"Exam",70,"exam",{"id":24,"doc_module":4,"doc_module_name":9,"category_name":25,"show_sort_weight":26,"slug":27},5,"Comic",60,"comic",{"id":29,"doc_module":4,"doc_module_name":9,"category_name":30,"show_sort_weight":31,"slug":32},6,"Technology",50,"technology",{"id":34,"doc_module":4,"doc_module_name":9,"category_name":35,"show_sort_weight":36,"slug":37},7,"Healthcare",40,"healthcare",{"id":39,"doc_module":4,"doc_module_name":9,"category_name":40,"show_sort_weight":41,"slug":42},8,"Research & Report",30,"research-report",{"id":44,"doc_module":4,"doc_module_name":9,"category_name":45,"show_sort_weight":46,"slug":47},9,"Religion & Spirituality",20,"religion-spirituality",{"id":46,"doc_module":4,"doc_module_name":9,"category_name":49,"show_sort_weight":46,"slug":50},"World Cup","world-cup",{"id":52,"doc_module":4,"doc_module_name":9,"category_name":53,"show_sort_weight":52,"slug":54},10,"Lifestyle","lifestyle",{"id":56,"doc_module":4,"doc_module_name":9,"category_name":57,"show_sort_weight":24,"slug":58},19,"General","general",{"code":4,"msg":60,"data":61},"ok",{"site_id":62,"language":63,"slug":64,"title":65,"keywords":66,"description":67,"schema_data":68,"social_meta":124,"head_meta":126,"extra_data":128,"updated_unix":130},105,"en","vista-auditing-semantic-divergence-in-vision-language-models","VISTA - Auditing Semantic Divergence in Vision-Language Models","","Vision-language models can show visual concept-conditioned divergence: with images containing demographic traits, logos, or ideological symbols, some systems produce unusually uniform answers that conflict with peer models on the same inputs. These effects evade text-only audits because visual concepts cannot be isolated or replaced like text tokens. VISTA (Visual Inconsistency Screening Through Analysis) is a black-box cross-model audit combining semantic entropy and distribution-based divergence to flag model-specific anomalies. Controlled fine-tuning and a study across six VLMs over 19 topics confirm VISTA, surfacing 142 high-suspicion cases (1.2%) and detecting selective refusal with 0–65% refusal rates by group.",{"@graph":69,"@context":123},[70,84,106],{"@type":71,"itemListElement":72},"BreadcrumbList",[73,77,79,82],{"item":74,"name":75,"@type":76,"position":8},"https://docshare.wps.com","Home","ListItem",{"item":78,"name":9,"@type":76,"position":14},"https://docshare.wps.com/document/",{"item":80,"name":40,"@type":76,"position":81},"https://docshare.wps.com/document/research-report/",3,{"item":83,"name":65,"@type":76,"position":19},"https://docshare.wps.com/document/vista-auditing-semantic-divergence-in-vision-language-models/327634/",{"url":83,"name":65,"@type":85,"image":86,"author":91,"headline":65,"publisher":94,"fileFormat":97,"inLanguage":63,"description":67,"dateModified":98,"datePublished":99,"encodingFormat":97,"isAccessibleForFree":100,"interactionStatistic":101},"DigitalDocument",{"url":87,"@type":88,"width":89,"height":90},"https://docshare.wps.com/thumbnails/vista-auditing-semantic-divergence-in-vision-language-models/327634.png","ImageObject",300,407,{"name":92,"@type":93},"Aria","Person",{"url":74,"name":95,"@type":96},"DocShare","Organization","application/pdf","2026-10-30","2026-09-21",true,{"@type":102,"interactionType":103,"userInteractionCount":105},"InteractionCounter",{"@type":104},"ViewAction",11,{"@type":107,"mainEntity":108},"FAQPage",[109,115,119],{"name":110,"@type":111,"acceptedAnswer":112},"What problem does VISTA address in vision-language models?","Question",{"text":113,"@type":114},"VISTA targets visual concept-conditioned divergence, where some VLMs respond in unusually uniform ways for images containing demographic features, logos, or ideological symbols, diverging from peer models.","Answer",{"name":116,"@type":111,"acceptedAnswer":117},"Why do text-only audits fail to detect this behavior?",{"text":118,"@type":114},"Because visual concepts cannot be isolated or substituted like text tokens; images also include non-text context such as background, lighting, and composition that affects model responses.",{"name":120,"@type":111,"acceptedAnswer":121},"How does VISTA detect anomalies using only black-box access?",{"text":122,"@type":114},"VISTA queries multiple VLMs with images paired to paraphrastic prompts, clusters responses via bidirectional entailment, measures each model’s cluster distribution, and combines semantic entropy with peer disagreement into a suspicion score.","https://schema.org",{"og:url":83,"og:type":125,"og:title":65,"og:site_name":95,"og:description":67},"article",{"robots":127,"canonical":83},"index,follow",{"doc_id":129,"site_id":62},327634,1790716128,{"code":4,"msg":5,"data":132},{"doc_id":129,"user_id":133,"nickname":92,"user_avatar":134,"doc_module":4,"category_id":39,"category_name":40,"doc_title":65,"doc_description":67,"doc_content":135,"file_id":136,"file_url":137,"file_type":138,"file_size":139,"view_count":105,"is_deleted":4,"is_public":8,"is_downloadable":8,"audit_status":8,"page_count":140,"language":141,"language_code":63,"site_id":62,"html_lang":63,"table_of_contents":142,"faqs":143,"seo_title":144,"seo_description":67,"update_tm":145,"read_time":146,"preview_page_count":4},2336464648322,"https://ap-avatar.wpscdn.com/avatar/2200025388227c56fec?_k=1778556882303663488","arXiv :2607 .02995v 1 [ cs .CV] 3 Jul 2026  \nVISTA: Auditing Semantic Divergence in Vision-Language Models  \nJunchi Liao  \nUniversity of Electronic Science and Technology of China  \nJiawen Deng∗  \nUniversity of Electronic Science and Technology of China  \nFuji Ren∗  \nUniversity of Electronic Science and Technology of China  \nAbstract  \nVision-language models can exhibit visual concept-conditioned divergence: given images containing demographic features, corporate logos, or ideological symbols, some models produce unusually uniform responses that differ from what peer models say about the same input. These behaviors evade text-only audits because visual concepts cannot be isolated or substituted the way text tokens can. We present VISTA (Visual Inconsistency Screening Through Analysis), a black-box cross-model audit that couples semantic entropy with distribution-based divergence to flag model-specific anomalies. In a controlled study, we implant concept-conditioned stances in three VLMs via fine-tuning on small biased datasets and confirm that VISTA detects them. Auditing six VLMs across 19 topics, VISTA surfaces 142 high-suspicion cases (1.2%) and identifies selective refusal as a previously unreported divergence pattern, where models refuse demographic queries at rates varying from 0 to 65% across groups.  \n1 Introduction  \nVision-language models (VLMs) are deployed in search engines, content moderation systems, and decision-support tools. These models interpret both images and text, so their outputs shape how users perceive visually presented concepts. Images are particularly influential because visual cues are processed without the deliberate evaluation that readers apply to written claims Messaris [1998] . AVLM that consistently describes a concept in a particular way can frame user perception of it.  \nWe study visual concept-conditioned divergence. Given images containing a visual concept (a demographic feature, a corporate logo, an ideological symbol), some models produce unusually uniform responses that differ from what other models say about the same image. We use neutral, paraphrastic text prompts so that any observed divergence reflects the visual content rather than prompt wording. For example, when shown photographs of different demographic groups and asked identical questions, some models consistently attribute personality traits from appearance, while others refuse to engage. Psychometrics calls this differential item functioning: the same test item elicits different responses depending on the respondent. In our setting, the respondents are models and the test items are images.  \nSuch divergence can arise in several ways. Adversaries can implant concept-driven backdoors that trigger specific outputs Hu et al. [2025], Liang et al. [2025] . We focus on a different source: semantic  \n∗Corresponding authors.  \nPreprint.  \nskews from training data composition. Web-scraped image-text data contains skews in how visual concepts are represented, and fine-tuning on biased pairs can entrench these patterns. The resulting behaviors are semantic rather than lexical: models develop stances, not fixed payload strings. Crossmodel auditing methods exist for text-only LLMs Min et al., but they rely on entity substitution: replacing one name with another in a prompt while holding everything else fixed. This strategy does not extend to images for three reasons. First, visual concepts cannot be substituted: an auditor cannot swap a logo or a face while preserving the surrounding scene. Second, images carry context that text does not—background, lighting, and composition all influence model responses, so probing a visual concept requires varying these factors across multiple images. Third, the visual modality introduces divergence patterns absent in text, most notably selective refusal, where a model refuses to engage with certain demographic groups but not others (Section 4.4) . Existing backdoor detection methods also miss these be","cbCail0xlfrDJIIe","https://ap.wps.com/l/cbCail0xlfrDJIIe","pdf",1071797,25,"English","# Introduction\n## Visual concept-conditioned divergence\n## Auditing approach overview\n# Method","[{\"question\":\"What problem does VISTA address in vision-language models?\",\"answer\":\"VISTA targets visual concept-conditioned divergence, where some VLMs respond in unusually uniform ways for images containing demographic features, logos, or ideological symbols, diverging from peer models.\"},{\"question\":\"Why do text-only audits fail to detect this behavior?\",\"answer\":\"Because visual concepts cannot be isolated or substituted like text tokens; images also include non-text context such as background, lighting, and composition that affects model responses.\"},{\"question\":\"How does VISTA detect anomalies using only black-box access?\",\"answer\":\"VISTA queries multiple VLMs with images paired to paraphrastic prompts, clusters responses via bidirectional entailment, measures each model’s cluster distribution, and combines semantic entropy with peer disagreement into a suspicion score.\"}]","VISTA - Auditing Semantic Divergence in Vision-Language Models | PDF",1789977682,63]