AI and ethics最新文献

{"title":"Residual governance for responsible embodied AI: authority, contestability, and cognitive sovereignty","authors":"Chia-Wei Sun","doi":"10.1007/s43681-026-01400-z","DOIUrl":"10.1007/s43681-026-01400-z","url":null,"abstract":"<div><p>Artificial intelligence (AI) ethics increasingly concerns systems that do not merely recommend but intervene: clinical decision support, human–machine teaming, robotics and automated handover protocols can shift timing, attention and authority under uncertainty. This article argues that responsible embodied AI requires residual governance: an explicit account of which approximation errors remain, which are safe to tolerate and who may act on them. The paper develops two linked concepts. A residual ledger records action-relevant remainders across physical, causal, human-state, latency, transfer and energy dimensions. An authority ledger records who converts residual evidence into intervention, on what basis, with what uncertainty and with what opportunity for contestation. This framework extends familiar themes in bounded rationality, active inference and safe reinforcement learning by treating authority, contestability and oversight cost as first-class ethical variables rather than external governance add-ons. It uses physical AI, physiological computing and human-AI handover as pressure cases, and proposes evaluation through action calibration, dual-task interference and auditable authority transitions. The central claim is that cognitive sovereignty is preserved not by eliminating all algorithmic influence, but by making influence accountable, contestable and reversible when AI systems act under approximation.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"6 5","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-09-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s43681-026-01400-z.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149034582","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"From hidden use to accountable governance: an ethical maturity model for shadow AI in organizations","authors":"Alfredo Gavito Romero,&nbsp;Virgilio López-Morales,&nbsp;Anilu Franco-Arcega,&nbsp;Manuel Alejandro Ojeda-Misses","doi":"10.1007/s43681-026-01413-8","DOIUrl":"10.1007/s43681-026-01413-8","url":null,"abstract":"<div><p>Shadow AI involves employees using public, personal, or other unauthorized artificial intelligence tools without having obtained formal approval or oversight from the organization. This article contends that Shadow AI is more than just a Cybersecurity or compliance issue; it is also an ethical governance problem because hidden use obscures accountability, reduces transparency, shifts responsibility onto individual workers, and gives rise to concerns regarding privacy, fairness, and the need for human oversight. By means of an integrative literature review and the construction of a conceptual model, the study draws together evidence from research on Shadow IT, studies concerning the use of generative artificial intelligence in the workplace, the AI ethics literature, and international governance frameworks. The patterns identified are converted into governance capabilities and are then arranged according to the degree of dependency to produce the Shadow AI Governance Maturity Model (SAGMM), a five-level framework that progresses from Invisible through to Visible, Controlled, Enabled, Adaptive governance. The model is conceptually assessed using evidence traceability, comparisons with existing frameworks, and examples from actual organizations, and it is also applied to anonymized data collected from an organization with 520 employees over a six-month period. The results of the organizational assessment show how the SAGMM indicators and the interpretation of maturity levels can be applied in real-world situations, although the model is not formally validated by this assessment. The one-year improvement figures are clearly presented as illustrative and non-predictive governance objectives rather than as actual measured outcomes or causal estimates. The ethical analysis is based on deontological, consequentialist, justice, and virtue approaches and explores the conflict between organizational visibility and employee privacy, surveillance, and autonomy. The key contribution of the SAGMM is a discovery-first approach combined with control and enablement for cases of partially hidden AI use by employees. The article maintains that organizations have a shared responsibility for establishing the conditions under which legitimate, governed AI use is both practical and proportionate and is more appealing than concealment.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"6 5","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-09-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149034833","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"From algorithmic performance to ethical governance: a deployment-oriented framework for responsible healthcare AI","authors":"Shehu Mohammed,&nbsp;Neha Malhotra,&nbsp;Sourabh Kumar","doi":"10.1007/s43681-026-01408-5","DOIUrl":"10.1007/s43681-026-01408-5","url":null,"abstract":"<div><p>Artificial intelligence (AI) is increasingly being integrated into healthcare for diagnosis, decision support, and operational optimization, yet real-world deployment remains inconsistent and institutionally challenging. This study presents a critical and integrative review of the ethical, regulatory, technical, and socio-technical issues shaping responsible healthcare AI implementation. Rather than proposing a new algorithm method, the study develops a governance-oriented conceptual synthesis that reframes deployment as a lifecycle integration challenge. Drawing on interdisciplinary literature and global regulatory frameworks, the review identifies recurring barriers, including dataset shift, performance drift, algorithmic bias, accountability gaps, and institutional misalignment. While these challenges are widely recognized, they are often treated as isolated concerns. The central contribution of this study is the proposed Lifecycle-Integrated Healthcare AI Governance Framework (LIHAG), a four-layer deployment-oriented governance model integrating technical robustness, ethical safeguards, regulatory compliance, and human oversight as interdependent dimensions of responsible AI deployment. Through illustrative healthcare use cases, the analysis demonstrates how lifecycle-integrated governance can support ethically aligned, regulation-ready, and institutionally sustainable AI implementation. By shifting evaluative emphasis from predictive accuracy alone to coordinated governance design, this study contributes a structured framework for responsible and sustainable healthcare AI deployment.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"6 5","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-09-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149034641","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Meaningful human authority in AI mediated decision making an authority vector and bottleneck principle","authors":"Connor Nitchals","doi":"10.1007/s43681-026-01414-7","DOIUrl":"10.1007/s43681-026-01414-7","url":null,"abstract":"<div><p>Human-in-the-loop requirements are often treated as evidence that an AI system remains under human control. Yet the placement of a person in a workflow does not establish that the person can understand, direct, or justify the resulting decision. Existing work identifies important pieces of the problem through meaningful human control, human factors, accountability, explainability, and procedural rights, but it does not provide a unified role-level test for whether human involvement amounts to authority in practice. This paper develops such a test. Meaningful human authority is defined as a relational and institutional capacity to govern an AI-mediated decision through seven necessary conditions: epistemic access, normative standing, temporal capacity, decisional independence, intervention power, contestability, and accountability traceability. The account introduces an authority vector and a context-specific threshold model. The minimum operator is used as a diagnostic of the most serious noncompensable shortfall, not as a measure of moral worth or a universally comparable score. The paper also clarifies that epistemic access requires practical interpretive competence. A reviewer must know enough about the underlying task to assess evidence, recognize error, and formulate an alternative judgment; institutions must preserve that competence against automation-related deskilling. A comparative benefits-eligibility case and cross-sector applications show how the model distinguishes nominal review from meaningful authority. The paper concludes with an Authority Impact Assessment, legal and regulatory implications, limitations, and a research agenda. The central claim is limited but demanding: when an institution attributes responsibility to a human reviewer, it must provide the information, competence, time, independence, power, challenge procedures, and responsibility structure needed for that reviewer to act as an authority rather than a ceremonial checkpoint.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"6 5","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-09-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149034830","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"From AI ethics principles to academic design practice: a faculty-led generative AI case in doctoral education","authors":"Scott A. Courtney,&nbsp;Bridget K. Mulvey","doi":"10.1007/s43681-026-01374-y","DOIUrl":"10.1007/s43681-026-01374-y","url":null,"abstract":"<div><p>This article examines how principles of responsible generative artificial intelligence (AI) use can inform faculty-led academic design in higher education. Using the development of the Doctoral Scholar Development Framework for Education (DSDF-E) as a design case, the article documents how generative AI supported organization, comparison, redundancy review, terminology refinement, and language revision while faculty retained responsibility for values, content, institutional fit, scholarly warrant, and final decisions. The case contributes to AI ethics by illustrating one institutional example of how transparency, accountability, human oversight, verification, contextualization, and bounded use can be enacted in program-level academic design. Rather than treating AI ethics as abstract principles, the article shows how ethical safeguards may be incorporated into everyday institutional work. The resulting descriptive model identifies seven phases of faculty-led AI-supported framework development and six practical design principles for responsible use: bounded use, provisionality, contextualization, verification, transparency, and data protection and confidentiality.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"6 5","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-09-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s43681-026-01374-y.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149034831","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Perspectives on roles and rewards in new cocreative systems for music-making","authors":"Tom Collins,&nbsp;Kyle Worrall,&nbsp;Chenyu Gao,&nbsp;Jemily Rime,&nbsp;Raina Murnak,&nbsp;Mark Hanslip,&nbsp;Ashay Dave","doi":"10.1007/s43681-026-01380-0","DOIUrl":"10.1007/s43681-026-01380-0","url":null,"abstract":"<div><p>Controversy and ethical concerns regarding data sourcing, copyright, and remuneration have emerged in relation to the application of generative artificial intelligence (AI) to creative tasks in writing, art, and music. The systems in which the AI tools are embedded lack sufficient evidence of thought or documentation regarding the roles and rewards for musicians (as data creators), model preppers, and model developers. The main contribution of this paper is to provide an examination of the essential components of such systems: who are the actors, how should they interact with one another, and how should they be charged/rewarded for their actions? Via (1) a user study involving ten professional producers, songwriters, and composers, and (2) reflections on our own careers as professional musicians, we find that with the exception of one system under consideration, called Cocreate, existing systems de-emphasize or disregard the role of a musician as data creator, but that this type of user can and should be rewarded for contributions to this new breed of music-making systems. An outcome of this paper is that we now have a blueprint via which we can bridge technology and musical creativity in ways that are more transparent and fair, while continuing to monitor perspectives of users and contributors.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"6 5","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-09-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s43681-026-01380-0.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149034640","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Autonomous intelligence in public trust: ethical frameworks for large language models in federal cybersecurity threat hunting","authors":"Adedamola Ajibola","doi":"10.1007/s43681-026-01335-5","DOIUrl":"10.1007/s43681-026-01335-5","url":null,"abstract":"<div><p>The integration of large language models (LLMs) into federal cybersecurity threat hunting promises significant advances in speed, scale, and analytical capability, particularly in environments characterized by high data volumes and persistent nation-state adversaries. As U.S. federal agencies confront increasingly sophisticated cyber threats, LLM-enabled systems offer the potential to augment human analysts through automated threat intelligence processing, cross-agency pattern recognition, and continuous monitoring of complex attack surfaces. However, the deployment of autonomous or semi-autonomous AI systems within national security contexts raises profound ethical and governance challenges that extend beyond traditional cybersecurity concerns. This paper examines the ethical tensions arising from hypothetical and emerging LLM deployment in federal threat hunting operations, focusing on conflicts between autonomy and human oversight, accuracy and civil liberties, transparency and operational security, and the allocation of responsibility when AI systems err. Using the SolarWinds supply chain compromise as an illustrative case study, the analysis demonstrates how LLM capabilities could both mitigate detection failures and introduce new risks if deployed without robust safeguards. The paper argues that effective integration of LLMs into federal cybersecurity requires an adaptive ethical governance framework grounded in human-centered design, proportionality, accountability, and democratic oversight. It advances a hybrid human-AI model that preserves human judgment in consequential decisions while leveraging AI for pattern recognition and data synthesis. By situating LLM-enabled threat hunting within broader debates on public trust and responsible innovation, this work contributes to the growing literature on AI ethics in high-stakes government and national security applications.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"6 5","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-09-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149034832","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Fairness versus profitability in automated machine learning","authors":"Paolo Giudici,&nbsp;Branka Hadji Misheva,&nbsp;Giacomo Piana","doi":"10.1007/s43681-026-01395-7","DOIUrl":"10.1007/s43681-026-01395-7","url":null,"abstract":"<div><p>We consider the application of automated machine learning models in a real credit lending platform, and empirically study whether the trade-off between fairness and profitability can be improved by means of fairness processing methods. Fairness is evaluated along the principle of Separation, while profitability is computed as a loan-level profit function, aggregated at the portfolio level. The empirical findings reveal that none of the considered fairness processors reduces bias without an increase in costs, in the form of higher false positive or of higher false negative rates.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"6 5","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-09-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s43681-026-01395-7.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149034895","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"AI resistance and epistemic agency: toward a relational framework for democratic and human-centered AI governance","authors":"Faisal Saeed Malik","doi":"10.1007/s43681-026-01396-6","DOIUrl":"10.1007/s43681-026-01396-6","url":null,"abstract":"<div><p>Artificial intelligence (AI) increasingly mediates communication, interpretation, decision-making, and knowledge production, raising questions about how computational systems acquire epistemic authority. At the same time, resistance to AI is emerging across educational, creative, infrastructural, professional, and civic settings. This conceptual paper examines what such resistance reveals about the distribution and contestability of epistemic authority in AI-mediated environments. Drawing on Science and Technology Studies, critical AI studies, AI ethics, political philosophy, and epistemic theory, it extends emerging accounts of <i>AI-mediated epistemic capture</i> by connecting them to <i>epistemic agency, resistance, and governance</i>. Epistemic capture is conceptualized as a gradual and relational process through which interpretive authority can become increasingly organized around computational systems and difficult to contest. The paper develops a typology encompassing quiet refusal, epistemic, infrastructural, educational, creative, and decolonial resistance, together with reclamation, while emphasizing that resistance is not inherently emancipatory or epistemically beneficial. Instead, resistance can operate diagnostically by revealing conditions under which epistemic agency becomes constrained. Building on this argument, the paper develops a <i>relational epistemic governance framework</i> integrating reflexivity, transparency, contestability, human adjudication, epistemic pluralism, democratic legitimacy, and sustainability. The framework shifts attention from governing AI systems alone toward the institutional organization of epistemic authority: who can interpret, challenge, adjudicate, and reshape AI-mediated knowledge, and under what conditions such authority remains democratically accountable.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"6 5","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-09-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s43681-026-01396-6.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149034892","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Developing nonviolent artificial intelligence: nonviolence as a principle of ethical AI","authors":"Andrew Fiala","doi":"10.1007/s43681-026-01399-3","DOIUrl":"10.1007/s43681-026-01399-3","url":null,"abstract":"<div><p>This paper considers how a principle of nonviolence could be integrated into the architecture of ethical AI. It considers how nonviolence could be deployed in the system prompts of autonomous vehicles (to prevent weaponization), in autonomous weapons systems, and in LLM’s employed as war room assistants. It explains how nonviolence differs from the principle of nonmaleficence as found in “principlism” in applied ethics in discussions of AI ethics. It shows how discussions of principled nonviolence are found in pacifism and the just war theory, while considering the idea of abolishing militaristic employment of AI systems. It concludes by imagining ways that nonviolent AI could push back against unethical usage of weapons systems, and how nonviolent AI could be useful in creating a more peaceful and less violent world.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"6 5","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-09-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149034467","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}