Disobedience for a cause: Leveraging Implicit Objectives in User Plans by LLM Surrogates

Disobedience for a cause: Leveraging Implicit Objectives in User Plans by LLM Surrogates | 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 update Disobedience for a cause: Leveraging Implicit Objectives in User Plans by LLM Surrogates Dale Peasley, Zachary Gray, Feyza M. Hafızoğlu, Sandip Sen This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9000614/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 6 You are reading this latest preprint version Abstract Large Language Models (LLM) are rapidly evolving from passive text generators into autonomous collaborators that must correctly infer what users really want and reliably execute complex tasks on their behalf. Yet current LLMs often misinterpret vague or under-specified instructions, fail to ask for clarification, or dispatch tools and executors in ways that are inconsistent with the user’s true goals, and may fail to meet user expectations, undermine user trust, and even result in serious execution errors. This article introduces LLM-powered ''disobedient'' agents that actively challenge ambiguous commands, infer user intent, and re-interpret user directives and requests, with the aim of closing the gap between what users mean and what agents do—thereby enhancing task execution correctness and user satisfaction. Disobedient agents can be useful when carefully designed to say ''no'' for justifiable reasons: preventing harm, correcting errors, and following more useful and effective alternative plans. This rational noncompliance can make AI systems more reliable, trustworthy, and resilient than purely obedient ones. This paper introduces the LLM-based Objective-informed Decision-making Architecture (LODA) to replace blind compliance with ''benevolent disobedience''. LODA alters the standard single-pass execution paradigm by separating AI decision-making into Objective Extraction and Plan Selection. Rather than directly executing a prompt, LODA infers the user's implicit primary goals and non-negotiable constraints before evaluating available actions against these extracted objectives. This architectural decoupling prevents LLMs from hallucinating justifications for false user claims and stops aspirational rhetoric from overriding baseline constraints. Empirical evaluations across multi-dimensional and information-asymmetric domains show that while innate sycophancy persists in scaled-up frontier models, LODA successfully mitigates these biases. These results demonstrate that deliberate, objective-driven defiance is essential for developing reliable, user-centric AI surrogates that prioritize underlying utility over literal directives. rebel agent intention recognition LLMs Full Text Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Reviewers agreed at journal 27 Apr, 2026 Reviewers agreed at journal 23 Apr, 2026 Reviewers invited by journal 20 Apr, 2026 Editor assigned by journal 02 Mar, 2026 Submission checks completed at journal 01 Mar, 2026 First submitted to journal 01 Mar, 2026 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-9000614","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"update","associatedPublications":[],"authors":[{"id":628403883,"identity":"24decb44-dee8-4d95-bef2-1bbfce2cd520","order_by":0,"name":"Dale Peasley","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA00lEQVRIiWNgGAWjYDCCwweYH36o+CcHZDIeAIscIKTlWAKbscSZA8YMbDDFRGhhkOBtO5DYQLQWvmO8Bwwk2+6kb7jfY3DwRw2DHN+NBPxaJI/xJTwoOPcsd8MxHoPDPMcYjCUJaTEAGm4gUcYM0QL0UOIGglqAKiV42JjTQYyDP/4x1BOppe1wAohxgLeNIcGACL+kAQM5zXDmsbSCw7x9EoYzzzzArwUYYoeBUWkjz3f48MaHP74BGccJ2MLAwIPCkyCkHFPLKBgFo2AUjAJMAACeqU6ef4P14AAAAABJRU5ErkJggg==","orcid":"","institution":"University of Tulsa","correspondingAuthor":true,"prefix":"","firstName":"Dale","middleName":"","lastName":"Peasley","suffix":""},{"id":628403885,"identity":"9f8d20ab-4042-4285-8567-50fd527e8dd8","order_by":1,"name":"Zachary Gray","email":"","orcid":"","institution":"University of Tulsa","correspondingAuthor":false,"prefix":"","firstName":"Zachary","middleName":"","lastName":"Gray","suffix":""},{"id":628403886,"identity":"d056c3a2-9cd6-49b3-930e-105f6483982a","order_by":2,"name":"Feyza M. Hafızoğlu","email":"","orcid":"","institution":"Istanbul Commerce University","correspondingAuthor":false,"prefix":"","firstName":"Feyza","middleName":"M.","lastName":"Hafızoğlu","suffix":""},{"id":628403887,"identity":"259da4c8-9865-48ad-ad5b-54f37df6eb75","order_by":3,"name":"Sandip Sen","email":"","orcid":"","institution":"University of Tulsa","correspondingAuthor":false,"prefix":"","firstName":"Sandip","middleName":"","lastName":"Sen","suffix":""}],"badges":[],"createdAt":"2026-03-01 09:54:48","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9000614/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9000614/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":108006464,"identity":"7bf9f881-ae00-4bc3-ba67-da98c70d62b1","added_by":"auto","created_at":"2026-04-28 12:55:42","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":599849,"visible":true,"origin":"","legend":"","description":"","filename":"LODALLMbasedObjectiveinformedDecisionmakingArchitecture.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9000614/v1_covered_a37ba4be-52e2-4b6d-b058-06c3a1a31393.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Disobedience for a cause: Leveraging Implicit Objectives in User Plans by LLM Surrogates","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"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":"autonomous-agents-and-multi-agent-systems","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"agnt","sideBox":"Learn more about [Autonomous Agents and Multi-Agent Systems](http://link.springer.com/journal/10458)","snPcode":"10458","submissionUrl":"https://submission.nature.com/new-submission/10458/3","title":"Autonomous Agents and Multi-Agent Systems","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"rebel agent, intention recognition, LLMs","lastPublishedDoi":"10.21203/rs.3.rs-9000614/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9000614/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":" Large Language Models (LLM) are rapidly evolving from passive text generators into autonomous collaborators that must correctly infer what users really want and reliably execute complex tasks on their behalf. Yet current LLMs often misinterpret vague or under-specified instructions, fail to ask for clarification, or dispatch tools and executors in ways that are inconsistent with the user’s true goals, and may fail to meet user expectations, undermine user trust, and even result in serious execution errors. This article introduces LLM-powered ''disobedient'' agents that actively challenge ambiguous commands, infer user intent, and re-interpret user directives and requests, with the aim of closing the gap between what users mean and what agents do—thereby enhancing task execution correctness and user satisfaction. Disobedient agents can be useful when carefully designed to say ''no'' for justifiable reasons: preventing harm, correcting errors, and following more useful and effective alternative plans. This rational noncompliance can make AI systems more reliable, trustworthy, and resilient than purely obedient ones. \nThis paper introduces the LLM-based Objective-informed Decision-making Architecture (LODA) to replace blind compliance with ''benevolent disobedience''. LODA alters the standard single-pass execution paradigm by separating AI decision-making into Objective Extraction and Plan Selection. Rather than directly executing a prompt, LODA infers the user's implicit primary goals and non-negotiable constraints before evaluating available actions against these extracted objectives. This architectural decoupling prevents LLMs from hallucinating justifications for false user claims and stops aspirational rhetoric from overriding baseline constraints. Empirical evaluations across multi-dimensional and information-asymmetric domains show that while innate sycophancy persists in scaled-up frontier models, LODA successfully mitigates these biases. These results demonstrate that deliberate, objective-driven defiance is essential for developing reliable, user-centric AI surrogates that prioritize underlying utility over literal directives.","manuscriptTitle":"Disobedience for a cause: Leveraging Implicit Objectives in User Plans by LLM Surrogates","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-04-27 12:37:16","doi":"10.21203/rs.3.rs-9000614/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"reviewerAgreed","content":"312228685065288301564768635162613452198","date":"2026-04-27T14:36:09+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"222080146468576423399425467491263071079","date":"2026-04-23T13:09:44+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-04-21T00:34:10+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-03-03T00:47:20+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-03-02T00:59:09+00:00","index":"","fulltext":""},{"type":"submitted","content":"Autonomous Agents and Multi-Agent Systems","date":"2026-03-01T09:43:29+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"autonomous-agents-and-multi-agent-systems","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"agnt","sideBox":"Learn more about [Autonomous Agents and Multi-Agent Systems](http://link.springer.com/journal/10458)","snPcode":"10458","submissionUrl":"https://submission.nature.com/new-submission/10458/3","title":"Autonomous Agents and Multi-Agent Systems","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"ad74f8eb-54f8-47da-8da0-d9e52a4503b3","owner":[],"postedDate":"April 27th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-04-27T12:37:16+00:00","versionOfRecord":[],"versionCreatedAt":"2026-04-27 12:37:16","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9000614","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9000614","identity":"rs-9000614","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}