Quantifying Hallucination Bias in AI-Generated Deepfakes: A Multimodal Analysis Using Divergence Metrics

Quantifying Hallucination Bias in AI-Generated Deepfakes: A Multimodal Analysis Using Divergence Metrics | 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 Quantifying Hallucination Bias in AI-Generated Deepfakes: A Multimodal Analysis Using Divergence Metrics Mohak Dwarkadhish Sharma, Anshita Bharadwaj This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6771530/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 The rapid development of artificial intelligence (AI) has transformed content creation, while also introducing new challenges, particularly with AI ’hallucinations’—instances where models generate incorrect or fabricated outputs. This study hypothesizes that hallucinations, often resulting from model over-fitting, can mimic or facilitate the generation of deepfakes. We propose a novel divergence metric θ to quantitatively differentiate hallucinated outputs from those produced by deepfake models. Leveraging the FaceForensics++ dataset and a dual-model training strategy using autoencoders, we contrast the behavior of a regularized deepfake model against an overfitted hallucination-prone model. Empirical evaluation using θ-distributions, classification metrics, and t-SNE visualization reveals measurable differences in output divergence. These findings provide insight into the ethical and technical implications of model hallucination, contributing toward more robust digital forensics and detection systems. AI Hallucinations Deepfakes Generative AI Machine Unlearning Ethical AI Digital Forensics θ-divergence Autoencoders FaceForensics++ Neural Overfitting Full Text Additional Declarations No competing interests reported. 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-6771530","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":471863938,"identity":"27bb3b7e-6652-459d-af7e-09316b6a4be3","order_by":0,"name":"Mohak Dwarkadhish Sharma","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAt0lEQVRIiWNgGAWjYFACHiA2kJDjh/CYidVSYGEs2UCalg8ViRsOEKtFt//swc8VBhLGxsePP5NgqLBObCCkxexGXrLkGaBfzM7kmEkwnEknRguPgWQD0BazAzlsEoxth4nQcv6M8U+glsTN/c+fSTD+I0bLgRwzkC2JGyQSzCQYG4jRciPHzBLkMIkbb4wtEo6lGxPlsJsNf+rk+PvTH974UGMtS1ALKkggTfkoGAWjYBSMAlwAAHWePP2+vD1YAAAAAElFTkSuQmCC","orcid":"","institution":"Arizona State University","correspondingAuthor":true,"prefix":"","firstName":"Mohak","middleName":"Dwarkadhish","lastName":"Sharma","suffix":""},{"id":471863939,"identity":"ab5c6d2b-7355-4ebe-8529-f14a26125790","order_by":1,"name":"Anshita Bharadwaj","email":"","orcid":"","institution":"Arizona State University","correspondingAuthor":false,"prefix":"","firstName":"Anshita","middleName":"","lastName":"Bharadwaj","suffix":""}],"badges":[],"createdAt":"2025-05-29 00:53:12","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6771530/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6771530/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":92034682,"identity":"91618e19-c16e-4f0e-8d08-bf8b060f7647","added_by":"auto","created_at":"2025-09-23 23:53:36","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":251805,"visible":true,"origin":"","legend":"","description":"","filename":"QuantifyingHallucinationBiasinAIGeneratedDeepfakesAMultimodalAnalysisUsingDivergenceMetrics.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6771530/v1_covered_0a17f81b-551d-499c-b3fd-3f0ee9e52deb.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Quantifying Hallucination Bias in AI-Generated Deepfakes: A Multimodal Analysis Using Divergence Metrics","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"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":"AI Hallucinations, Deepfakes, Generative AI, Machine Unlearning, Ethical AI, Digital Forensics, θ-divergence, Autoencoders, FaceForensics++, Neural Overfitting","lastPublishedDoi":"10.21203/rs.3.rs-6771530/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6771530/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThe rapid development of artificial intelligence (AI) has transformed content creation, while also introducing new challenges, particularly with AI ’hallucinations’—instances where models generate incorrect or fabricated outputs. This study hypothesizes that hallucinations, often resulting from model over-fitting, can mimic or facilitate the generation of deepfakes. We propose a novel divergence metric θ to quantitatively differentiate hallucinated outputs from those produced by deepfake models. Leveraging the FaceForensics++ dataset and a dual-model training strategy using autoencoders, we contrast the behavior of a regularized deepfake model against an overfitted hallucination-prone model. Empirical evaluation using θ-distributions, classification metrics, and t-SNE visualization reveals measurable differences in output divergence. These findings provide insight into the ethical and technical implications of model hallucination, contributing toward more robust digital forensics and detection systems.\u003c/p\u003e","manuscriptTitle":"Quantifying Hallucination Bias in AI-Generated Deepfakes: A Multimodal Analysis Using Divergence Metrics","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-06-18 13:00:09","doi":"10.21203/rs.3.rs-6771530/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":"7c49bdcc-0916-4544-b31b-8b5e99b20d45","owner":[],"postedDate":"June 18th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-09-23T23:53:27+00:00","versionOfRecord":[],"versionCreatedAt":"2025-06-18 13:00:09","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6771530","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6771530","identity":"rs-6771530","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}