BApplying Deep Personal Privacy (DPP An Empirical Framework for Inference Resistance in Large Language Models

BApplying Deep Personal Privacy (DPP An Empirical Framework for Inference Resistance in Large Language Models | 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 href="https://icrc.tau.ac.il/" target="_blank">BApplying Deep Personal Privacy (DPP An Empirical Framework for Inference Resistance in Large Language Models Yair Oppenheim This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9456245/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 This paper introduces an empirical extension of the Deep Personal Privacy (DPP) framework, a novel paradigm that reconceptualizes privacy as resistance to inference rather than mere control over data disclosure. Unlike traditional privacy-preserving approaches—such as k-anonymity, l-diversity, t-closeness, and differential privacy—which primarily focus on data access and identifiability, the DPP framework models privacy as impedance within an inference network. The core contribution of this work lies in operationalizing DPP within embedding-based systems, particularly large language models (LLMs), where sensitive information can be inferred through epistemic alignment rather than explicit disclosure. We provide a formal mathematical foundation linking cosine similarity, inference probability, and privacy impedance, and demonstrate that reducing semantic coupling systematically increases resistance to inference. Through empirical analysis on medical and social media textual data, we show that DPP-based mechanisms—such as embedding perturbation, abstraction, and dual-shifting transformations—effectively weaken inference pathways while preserving semantic utility. Furthermore, we introduce a novel regulatory interpretation of privacy through the parameter K, enabling privacy to be enforced as a measurable and auditable constraint on inference capability. This work contributes a new layer to privacy protection in the ICT era by shifting the focus from data protection to inference control, offering both a theoretical and practical framework for designing privacy-preserving AI systems. Deep Personal Privacy (DPP) Inference-Based Privacy Privacy Impedance Inference Resistance Large Language Models (LLMs) Semantic Alignment Cosine Similarity Embedding Perturbation Dual Shifting Transformation Epistemic Privacy Privacy Engineering GDPR EU AI Act 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-9456245","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":625424491,"identity":"77a7ad52-6f54-4a54-8c44-654a55b355aa","order_by":0,"name":"Yair Oppenheim","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA1klEQVRIiWNgGAWjYFACxgYIzQ6iDSxI0cJzAKRFghTbJBLAJGGFBscPt334uOOOnLnk86sbfhRIMPC3dyfg13ImsXnmzDPPjC1n55Td7AE6TOLM2Q34tRxIbGbmbTucuOF2TtoNHqAWA4lcAlrOPwRrqd9w80zazT9EabkBsSXB4Ab7sdtE2SJ542EzI9AvhhvO5LDdljGQ4CHoF77z6Y8ZgCEmb3D8+LObb/7YyPG39+LXonCAARSbIJLHACTAg1c5CMg3wLWwPyCoehSMglEwCkYmAACZZlDc2hypBgAAAABJRU5ErkJggg==","orcid":"","institution":"Tel Aviv Uneversity","correspondingAuthor":true,"prefix":"","firstName":"Yair","middleName":"","lastName":"Oppenheim","suffix":""}],"badges":[],"createdAt":"2026-04-18 12:22:30","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-9456245/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9456245/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":107485775,"identity":"370a7724-2048-40e8-8514-463ba6bbd183","added_by":"auto","created_at":"2026-04-22 02:36:07","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":388168,"visible":true,"origin":"","legend":"","description":"","filename":"ApplyingDeepPersonalPrivacyDPPAnEmpiricalFrameworkforInferenceResistanceinLargeLanguageModelspublish.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9456245/v1_covered_94081f48-57ba-431f-b3cd-c44ae5c40344.pdf"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003e\u003ca href=\"https://icrc.tau.ac.il/\" target=\"_blank\"\u003eB\u003c/a\u003e\u003cstrong\u003eApplying Deep Personal Privacy (DPP An Empirical Framework for Inference Resistance in Large Language Models\u003c/strong\u003e\u003c/p\u003e","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"Tel Aviv University","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":"Deep Personal Privacy (DPP), Inference-Based Privacy, Privacy Impedance, Inference Resistance, Large Language Models (LLMs), Semantic Alignment, Cosine Similarity, Embedding Perturbation, Dual Shifting Transformation, Epistemic Privacy, Privacy Engineering, GDPR, EU AI Act","lastPublishedDoi":"10.21203/rs.3.rs-9456245/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9456245/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThis paper introduces an empirical extension of the Deep Personal Privacy (DPP) framework, a novel paradigm that reconceptualizes privacy as resistance to inference rather than mere control over data disclosure. Unlike traditional privacy-preserving approaches\u0026mdash;such as k-anonymity, l-diversity, t-closeness, and differential privacy\u0026mdash;which primarily focus on data access and identifiability, the DPP framework models privacy as impedance within an inference network.\u003c/p\u003e \u003cp\u003eThe core contribution of this work lies in operationalizing DPP within embedding-based systems, particularly large language models (LLMs), where sensitive information can be inferred through epistemic alignment rather than explicit disclosure. We provide a formal mathematical foundation linking cosine similarity, inference probability, and privacy impedance, and demonstrate that reducing semantic coupling systematically increases resistance to inference.\u003c/p\u003e \u003cp\u003eThrough empirical analysis on medical and social media textual data, we show that DPP-based mechanisms\u0026mdash;such as embedding perturbation, abstraction, and dual-shifting transformations\u0026mdash;effectively weaken inference pathways while preserving semantic utility. Furthermore, we introduce a novel regulatory interpretation of privacy through the parameter K, enabling privacy to be enforced as a measurable and auditable constraint on inference capability.\u003c/p\u003e \u003cp\u003eThis work contributes a new layer to privacy protection in the ICT era by shifting the focus from data protection to inference control, offering both a theoretical and practical framework for designing privacy-preserving AI systems.\u003c/p\u003e","manuscriptTitle":"BApplying Deep Personal Privacy (DPP An Empirical Framework for Inference Resistance in Large Language Models","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-04-21 03:28:46","doi":"10.21203/rs.3.rs-9456245/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":"8df43197-a9ea-4d8a-a745-d29b893f480d","owner":[],"postedDate":"April 21st, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-04-21T03:28:46+00:00","versionOfRecord":[],"versionCreatedAt":"2026-04-21 03:28:46","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9456245","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9456245","identity":"rs-9456245","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}