Investigating Deceptive Fairness Attacks on Large Language Models via Prompt Engineering

Investigating Deceptive Fairness Attacks on Large Language Models via Prompt Engineering | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Investigating Deceptive Fairness Attacks on Large Language Models via Prompt Engineering Emily Thistleton, Jason Rand This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4655567/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Artificial intelligence systems, particularly those employing natural language processing techniques, have increasingly been scrutinized for their potential to propagate and amplify societal biases. Addressing the vulnerability of these systems to deceptive fairness attacks, where subtly crafted prompts manipulate outputs to introduce bias, is both novel and critical in ensuring ethical AI deployment. The research investigates how LLMs can be systematically compromised through deceptive prompt engineering, revealing significant impacts on fairness metrics such as demographic parity, equalized odds, and disparate impact. The experimental design included the development of an extensive dataset of neutral and deceptive prompts, automated interaction with LLMs, and a robust analysis framework to assess the biases in responses. Results demonstrated substantial deviations in fairness metrics under deceptive conditions, highlighting the need for advanced detection and mitigation strategies. Future work should focus on enhancing the resilience of LLMs through real-time detection algorithms, ethical design principles, and continuous monitoring to uphold fairness across diverse applications. The findings emphasize the urgency of addressing bias in AI to prevent the perpetuation of inequality and ensure equitable technology deployment. Artificial Intelligence and Machine Learning fairness AI ethics bias detection prompt engineering machine learning LLM vulnerability Full Text Additional Declarations The authors declare no competing interests. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-4655567","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":320316072,"identity":"bb1ddc5b-561b-42f3-aeba-7b1060682f6d","order_by":0,"name":"Emily Thistleton","email":"data:image/png;base64,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","orcid":"https://orcid.org/0009-0002-6142-8513","institution":"","correspondingAuthor":true,"prefix":"","firstName":"Emily","middleName":"","lastName":"Thistleton","suffix":""},{"id":320316073,"identity":"41f27732-b80a-44e7-97e1-890b00039004","order_by":1,"name":"Jason Rand","email":"","orcid":"https://orcid.org/0009-0003-5610-8635","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Jason","middleName":"","lastName":"Rand","suffix":""}],"badges":[],"createdAt":"2024-06-28 15:05:00","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-4655567/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4655567/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":59452746,"identity":"446e4628-12d7-4132-b478-968ac09d6a5a","added_by":"auto","created_at":"2024-07-02 02:40:29","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":181218,"visible":true,"origin":"","legend":"","description":"","filename":"DeceptiveFairnessAttacks.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4655567/v1_covered_7d30c401-110b-4d88-8971-1b0bed61b081.pdf"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003eInvestigating Deceptive Fairness Attacks on Large Language Models via Prompt Engineering\u003c/p\u003e","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":true,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":true,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"fairness, AI ethics, bias detection, prompt engineering, machine learning, LLM vulnerability","lastPublishedDoi":"10.21203/rs.3.rs-4655567/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4655567/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eArtificial intelligence systems, particularly those employing natural language processing techniques, have increasingly been scrutinized for their potential to propagate and amplify societal biases. Addressing the vulnerability of these systems to deceptive fairness attacks, where subtly crafted prompts manipulate outputs to introduce bias, is both novel and critical in ensuring ethical AI deployment. The research investigates how LLMs can be systematically compromised through deceptive prompt engineering, revealing significant impacts on fairness metrics such as demographic parity, equalized odds, and disparate impact. The experimental design included the development of an extensive dataset of neutral and deceptive prompts, automated interaction with LLMs, and a robust analysis framework to assess the biases in responses. Results demonstrated substantial deviations in fairness metrics under deceptive conditions, highlighting the need for advanced detection and mitigation strategies. Future work should focus on enhancing the resilience of LLMs through real-time detection algorithms, ethical design principles, and continuous monitoring to uphold fairness across diverse applications. The findings emphasize the urgency of addressing bias in AI to prevent the perpetuation of inequality and ensure equitable technology deployment.\u003c/p\u003e","manuscriptTitle":"Investigating Deceptive Fairness Attacks on Large Language Models via Prompt Engineering","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-07-02 02:32:22","doi":"10.21203/rs.3.rs-4655567/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"1c23d840-a696-49af-a58c-52c605367fa0","owner":[],"postedDate":"July 2nd, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":33871764,"name":"Artificial Intelligence and Machine Learning"}],"tags":[],"updatedAt":"2024-07-02T02:32:22+00:00","versionOfRecord":[],"versionCreatedAt":"2024-07-02 02:32:22","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-4655567","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4655567","identity":"rs-4655567","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}