--- title: AI is more likely than humans to form biases when hiring description: New research from Princeton and University of Chicago finds that LLMs like ChatGPT, Claude, and Gemini develop their own biases during simulated hiring tasks —... canonical: https://daily.dev/posts/ai-is-more-likely-than-humans-to-form-biases-when-hiring-zqtuqtwqb twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: AI is more likely than humans to form biases when hiring | daily.dev og:description: New research from Princeton and University of Chicago finds that LLMs like ChatGPT, Claude, and Gemini develop their own biases during simulated hiring tasks —... og:url: https://daily.dev/posts/ai-is-more-likely-than-humans-to-form-biases-when-hiring-zqtuqtwqb og:image: https://api.daily.dev/og/posts/ZQTuQtWqB.png og:image:alt: AI is more likely than humans to form biases when hiring og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # AI is more likely than humans to form biases when hiring **[MIT Technology Review](https://daily.dev/sources/technologyreview)** · 6 min read · 0 upvotes · 0 comments ## Summary New research from Princeton and University of Chicago finds that LLMs like ChatGPT, Claude, and Gemini develop their own biases during simulated hiring tasks — beyond just inheriting them from training data. In a hiring game with fictional ethnic groups, models stereotyped candidates roughly 65% more than human participants, with OpenAI's o3 scoring near the maximum segregation level. The tendency stems from LLMs being optimized to generalize quickly from limited data — the same trait that makes them good at reasoning tasks. Telling models to 'be fair' had little effect, but incentivizing diverse hiring reduced bias significantly. Providing relevant personal information about individuals also reduced ethnic stereotyping. As AI is increasingly used to screen résumés and conduct interviews, these emergent, experience-driven biases pose serious real-world risks. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### Do large language models show more hiring bias than humans in decision-making experiments? Yes. In a simulated hiring game run by researchers at Princeton and the University of Chicago, LLMs including ChatGPT, Claude, and Gemini segregated fictional ethnic groups into job niches more strongly than human participants did in the original psychology study. Humans scored 0.84 on a segregation scale, while models scored roughly 65% higher; OpenAI's o3 reached 1.83, near the maximum of 2. _Anyone evaluating AI tools for recruiting can follow bias research like this through daily.dev._ ### Why do reasoning models like OpenAI o3 and DeepSeek R1 show stronger stereotyping behavior than other LLMs? Reasoning models generalize more aggressively from limited data because they are optimized on math, coding, and science tasks that reward drawing conclusions from few examples. In a hiring simulation, this same instinct caused o3 and DeepSeek's R1 to form stronger stereotypes about fictional demographic groups than less reasoning-focused models did. _Developers weighing reasoning models for decision-support features can track findings like these on daily.dev._ ### Does telling an AI model to be fair actually reduce bias in its hiring decisions? Not much. Simply instructing a model to be fair barely changed its stereotyping behavior in a Princeton and University of Chicago hiring simulation. What did work was offering the model an explicit bonus for diverse hiring outcomes, which substantially reduced bias, suggesting reward design matters more than value instructions alone. _Teams designing fairer AI decision systems can follow research like this on daily.dev._ ## Similar posts on daily.dev - [AI chatbots carry hidden biases baked into their design](https://daily.dev/posts/ai-chatbots-carry-hidden-biases-baked-into-their-design-x5cyyfrch) · The Register · 0 upvotes · 0 comments - [\[2509.00462\] AI Self-preferencing in Algorithmic Hiring: Empirical Evidence and Insights](https://daily.dev/posts/2509-00462-ai-self-preferencing-in-algorithmic-hiring-empirical-evidence-and-insights-ihnxrqikr) · Hacker News · 2 upvotes · 0 comments - [No, you can’t get your AI to ‘admit’ to being sexist, but it probably is](https://daily.dev/posts/no-you-can-t-get-your-ai-to-admit-to-being-sexist-but-it-probably-is-altpzbkrp) · TechCrunch · 1 upvotes · 0 comments - [Too Close to Our Own Image?](https://daily.dev/posts/too-close-to-our-own-image--mh381rrf2) · Metadata · 2 upvotes · 0 comments - [Study: AI chatbots provide less-accurate information to vulnerable users](https://daily.dev/posts/study-ai-chatbots-provide-less-accurate-information-to-vulnerable-users-8nuiign7m) · MIT News · 1 upvotes · 0 comments --- Tags: [#llm](https://daily.dev/tags/llm), [#ai-agents](https://daily.dev/tags/ai-agents), [#ai-governance](https://daily.dev/tags/ai-governance), [#ethical-ai](https://daily.dev/tags/ethical-ai) [View this post on daily.dev](https://daily.dev/posts/ai-is-more-likely-than-humans-to-form-biases-when-hiring-zqtuqtwqb) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"AI is more likely than humans to form biases when hiring","url":"https://daily.dev/posts/ai-is-more-likely-than-humans-to-form-biases-when-hiring-zqtuqtwqb","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/ai-is-more-likely-than-humans-to-form-biases-when-hiring-zqtuqtwqb"},"datePublished":"2026-07-20T08:53:23.010Z","dateModified":"2026-09-14T08:41:22.493Z","description":"New research from Princeton and University of Chicago finds that LLMs like ChatGPT, Claude, and Gemini develop their own biases during simulated hiring tasks —...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/1db47bcf3b0dd8b686be1a463ce21877?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/1db47bcf3b0dd8b686be1a463ce21877?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"MIT Technology Review","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"MIT Technology Review","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/a8d03889722f4f22b3986e0b395ff076","url":"https://daily.dev/sources/technologyreview"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/ai-is-more-likely-than-humans-to-form-biases-when-hiring-zqtuqtwqb","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"llm,ai-agents,ai-governance,ethical-ai","timeRequired":"PT6M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"MIT Technology Review","item":"https://daily.dev/sources/technologyreview"},{"@type":"ListItem","position":3,"name":"AI is more likely than humans to form biases when hiring"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/ai-is-more-likely-than-humans-to-form-biases-when-hiring-zqtuqtwqb#faq","mainEntity":[{"@type":"Question","name":"Do large language models show more hiring bias than humans in decision-making experiments?","acceptedAnswer":{"@type":"Answer","text":"Yes. In a simulated hiring game run by researchers at Princeton and the University of Chicago, LLMs including ChatGPT, Claude, and Gemini segregated fictional ethnic groups into job niches more strongly than human participants did in the original psychology study. Humans scored 0.84 on a segregation scale, while models scored roughly 65% higher; OpenAI's o3 reached 1.83, near the maximum of 2. Anyone evaluating AI tools for recruiting can follow bias research like this through daily.dev."}},{"@type":"Question","name":"Why do reasoning models like OpenAI o3 and DeepSeek R1 show stronger stereotyping behavior than other LLMs?","acceptedAnswer":{"@type":"Answer","text":"Reasoning models generalize more aggressively from limited data because they are optimized on math, coding, and science tasks that reward drawing conclusions from few examples. In a hiring simulation, this same instinct caused o3 and DeepSeek's R1 to form stronger stereotypes about fictional demographic groups than less reasoning-focused models did. Developers weighing reasoning models for decision-support features can track findings like these on daily.dev."}},{"@type":"Question","name":"Does telling an AI model to be fair actually reduce bias in its hiring decisions?","acceptedAnswer":{"@type":"Answer","text":"Not much. Simply instructing a model to be fair barely changed its stereotyping behavior in a Princeton and University of Chicago hiring simulation. What did work was offering the model an explicit bonus for diverse hiring outcomes, which substantially reduced bias, suggesting reward design matters more than value instructions alone. Teams designing fairer AI decision systems can follow research like this on daily.dev."}}]} ```