The research, conducted by Sihan Lu, Yan Zhou, and Tao Hong of the Harbin Institute of Technology in China, starts from a theoretical observation about the academic literature. Most scholarship on generative AI in the public sector has examined the technology through a siloed lens, focusing on a single role at a time. Studies of AI as a governance tool emphasize efficiency gains and service improvement. Studies of AI as a regulatory object emphasize risk, liability, and the need for oversight. A smaller body of work treats AI as a novel decision-making agent with its own capacity to act. Each perspective is useful, but each is incomplete, and the fragmentation obscures what happens at the micro level when these roles intersect inside a single administrative system.

To overcome that limitation, the authors reconceptualize the public management system as a complex adaptive system, a framework borrowed from the study of ecosystems, markets, and other environments in which many interacting agents produce outcomes no single actor intends. Within this framework, generative AI is modeled as an adaptive agent rather than a static instrument. That shift matters because adaptive agents learn, respond to feedback, and alter the behavior of the other agents around them. A chatbot that answers citizen queries changes how civil servants work; a drafting assistant changes how policy documents are produced; a predictive model changes which cases get attention. The technology does not sit outside the governance system. It becomes part of the system’s own dynamics.

The empirical core of the study is a single-case analysis of Singapore’s Smart Nation initiative, one of the world’s most sustained and comprehensive national programs for embedding digital and intelligent technologies in public administration. By tracking governance strategies and their outcomes across a range of situations within that program, the researchers were able to observe how the same underlying technology was assigned different roles in different contexts, and how those assignments generated tensions that no single regulatory rule could resolve.

The central theoretical contribution of the paper is what the authors call the Governance Impossible Triangle. According to this framework, generative AI’s triple identity creates persistent tensions among three goals that public administrators must pursue at the same time: maximizing efficiency, safeguarding the legitimacy of decisions, and maintaining effective risk control. The triangle is impossible in the same sense that classic policy trilemmas are impossible. Push hard on efficiency, for example by automating more decisions end to end, and legitimacy comes under strain because citizens and officials lose the ability to understand and contest how outcomes were reached. Push hard on legitimacy, for example by requiring extensive human review and explanation, and the efficiency gains that justified adoption begin to erode. Push hard on risk control, and both efficiency and responsiveness suffer. The three corners pull against one another continuously, and no configuration satisfies all of them at once.

What makes the triangle particularly difficult is that the three identities of generative AI activate the three corners in different combinations. As a tool, the technology is primarily an efficiency asset, and the governance question is how much discretion to delegate to it. As a regulatory object, it is primarily a risk to be contained, and the question is how to oversee systems whose behavior even their developers cannot fully predict. As a decision-making agent, it touches legitimacy directly, because democratic governance rests on the premise that consequential decisions are attributable to accountable human institutions. When a single platform shifts between these roles depending on the task at hand, the governance burden shifts with it, often faster than rules and oversight structures can adapt.

The authors’ response to this dilemma is the Contextual Identity Configuration Model, a framework that rejects one-size-fits-all regulation in favor of dynamic, context-sensitive assignment of AI’s role. The model is driven by four variables: problem complexity, risk severity, value conflicts, and explainability requirements. Simple, low-risk, low-conflict problems with clear standards can tolerate a high degree of algorithmic tool-intensity, meaning the AI acts with substantial autonomy as an efficiency instrument. As complexity rises, as stakes rise, as values come into conflict, or as the need for explanation grows, the model calls for recalibrating the system, reducing agentic power and increasing human oversight, so that the technology reverts toward a supportive tool role rather than an autonomous one.

To make this concrete, the study analyzes four prototypical scenarios of AI deployment in public governance. In routine optimization, where tasks are standardized and risks are low, generative AI can operate with high tool-intensity, delivering efficiency with minimal threat to legitimacy. In professional assistance, the AI supports trained experts who retain judgment and accountability, a middle configuration that captures productivity while keeping humans in the loop. In high-stakes discretion, where decisions affect rights, liberties, or significant resources, the model demands strong explainability and human authority, constraining the AI to an advisory function. In complex crises, where speed matters but uncertainty is extreme, the framework calls for layered defenses and continuous feedback so that algorithmic recommendations inform but never replace accountable human command. Across all four scenarios, the governing principle is the same: management systems must dynamically recalibrate algorithmic tool-intensity and agentic power through continuous feedback loops rather than fixing the technology’s role in advance.

The practical implications the authors draw are institutional as much as technical. They argue that governments should build multi-layered defense systems, in which technical safeguards, procedural review, and human accountability operate together rather than as substitutes for one another. They also propose joint accountability networks, arrangements in which responsibility for AI-assisted decisions is distributed across developers, agencies, and oversight bodies in ways that remain legible to the public. The goal, in their formulation, is a profound alignment between technological empowerment and core public values, so that the efficiency of generative AI does not come at the cost of the legitimacy and trust on which public institutions depend.

The study arrives at a moment when many governments are experimenting with generative AI faster than they are theorizing it. Its contribution is to name the structural problem: the same technology that makes administration faster also makes it harder to explain, and the same flexibility that makes it useful makes it hard to regulate. By modeling the public sector as a complex adaptive system and the AI within it as an adaptive agent, the researchers offer policymakers a vocabulary for the trade-offs they are already making, often implicitly. The Governance Impossible Triangle will not be dissolved, the authors suggest, but it can be managed, provided that regulators abandon uniform rules in favor of an agile, context-based governance paradigm in which the identity of the technology, and the controls attached to it, are deliberately configured to match the demands of each task.

Subject of Research: The role and governance of generative AI in public administration and government decision-making

Article Title: The “Triple identity” of generative AI in public governance: dilemmas and adaptive solutions

Article References: Lu, S., Zhou, Y., & Hong, T. (2026). The “Triple identity” of generative AI in public governance: dilemmas and adaptive solutions. Humanities and Social Sciences Communications. https://doi.org/10.1057/s41599-026-09400-4

Image Credits: AI Generated

DOI: 10.1057/s41599-026-09400-4

Keywords: generative AI, public governance, complex adaptive systems, Smart Nation Singapore, algorithmic accountability, AI regulation, governance trilemma, decision-making agents, public administration, risk control, legitimacy, adaptive governance

The conceptual starting point is that legitimacy in global governance has never rested on sovereign command. International institutions hold authority only insofar as their justificatory claims are recognised, contested or denied by audiences around the world. Zürn’s idea of reflexive legitimation conflict describes a situation in which disputes escalate beyond questions of performance or participation to the meta-criteria of legitimacy itself: not only what counts as an effective basis for legitimation, but who decides this and how the answers get institutionalised. Pegram shows that AI governance, fragmented and weakly institutionalised, is unusually exposed to this kind of conflict, because there is no settled regime whose performance can be evaluated. What is contested instead is the evaluative script of global rule itself.

The empirical pivot is the rupture between the Biden and Trump approaches to AI. In 2023, Biden’s Executive Order 14110 on safe and trustworthy AI built explicitly on emerging multilateral precedents, including the OECD AI Principles of 2019, UNESCO’s 2021 Recommendation on the Ethics of AI adopted by 193 member states, and the European Commission’s risk-based AI Act. The order committed the United States to allied engagement in multilateral bodies, international technical standards and global AI safety monitoring, and even adopted the OECD’s definition of AI. Washington subsequently joined the 2023 Bletchley Declaration, signed the Council of Europe Framework Convention on AI in 2024, and sponsored a UN General Assembly resolution calling for safe, secure and trustworthy AI systems for sustainable development.

That trajectory reversed abruptly. On his first day back in office, Trump revoked the Biden order, leaving the United States without federal regulatory guidance. Three days later he issued Executive Order 14179, declaring national policy to sustain and enhance America’s global AI dominance in the name of human flourishing, economic competitiveness and national security. At the Paris AI Action Summit in February 2025, Vice President J.D. Vance warned that excessive regulation could kill a transformative industry, rejected the European model as burdensome, and framed international cooperation chiefly as a defence against authoritarian powers capturing technological infrastructure. The administration declined to sign the summit’s declaration on inclusive and sustainable AI, and in July 2025 withdrew from UNESCO, the body anchoring global AI ethics norms.

Crucially, the study finds, the Trump administration did not abandon international forums altogether. Instead it recast engagement in plurilateral and mercantile terms. America’s AI Action Plan, published in July 2025, promotes plurilateral controls on the AI technology stack while avoiding multilateral treaty bodies, coordinates with allies to extend US export controls against adversaries, and pledges to leverage the US position in standard-setting bodies such as the ITU, OECD, G7 and G20 to counter Chinese influence and ensure AI governance reflects American values. In August 2025, Washington backed a second term for Doreen Bogdan-Martin as ITU Secretary-General, explicitly linking that contest to control over the more than 150 AI standards already set. Trump framed the stakes geopolitically, warning that foreign nations must not make rules and regulations that make it impossible for American firms to do business.

China’s response forms the other half of the analysis, and it is here that the study identifies a striking role reversal. Beijing’s 2017 New Generation Artificial Intelligence Development Plan had already pledged active participation in global AI governance, but Western observers long dismissed Chinese claims as authoritarian and censorial, describing ambitions ranging from the anodyne to the dystopian. Successive initiatives, including a 2022 position paper on ethical AI governance and the Global AI Governance Initiative launched by Xi Jinping at the Third Belt and Road Forum in October 2023, emphasised sovereignty, developing AI for good, greater voice for developing countries and a UN-centred governance body. Yet these claims struggled to gain traction, and China’s inclusion in the Bletchley Declaration was contested so vigorously that UK Prime Minister Rishi Sunak had to defend it publicly.

Trump’s return marked the inflection point. Confronted with tightened chip sanctions in January 2025, Beijing did not retreat into secrecy as many predicted. Instead, firms such as Alibaba and DeepSeek released advanced models as open source, with Premier Li Qiang declaring that China’s innovation is open and open source and that the country was willing to share indigenous technologies. The strategy served a dual purpose: it mitigated the effect of sanctions by globalising Chinese models, and it recast China’s legitimacy claims, transforming a state once portrayed as censorial into a putative provider of AI as a global public good. At the World AI Conference in Shanghai in July 2025, Li called global AI governance fragmented, warned against technological monopoly, and unveiled a Global AI Governance Action Plan proposing two UN-based mechanisms: an International Scientific Panel on AI and a Global Dialogue on AI Governance.

The juxtaposition of the two strategies reveals what Pegram calls rival scripts of legitimation, each mobilising distinct repertoires conditioned by material power and normative legacy. The American script invokes sovereign discretion, market dynamism and technological nationalism, delegitimising liberal norms of ethical oversight and multilateral coordination while instrumentalising international forums for competitive advantage. The Chinese script invokes sovereign equality, developmental inclusion and UN-centred ethics, aligning open-sourcing practices with discursive claims of ethical multilateralism. By August 2025, Western commentators were warning of America’s AI colonialism, while Trump officials confirmed the United States was explicitly rejecting China’s call for global governance, offering instead customised AI packages tied to American technology. Neither script has achieved uncontested recognition, but their antagonism demonstrates that legitimacy itself has become the object of the contest.

The study’s findings carry implications well beyond AI. Legitimation, it concludes, is processual and precarious, continually reconfigured through cycles of challenge and response, in which delegitimation by one actor creates openings for others to reformulate their claims. Repertoires of legitimation remain structurally conditioned: US claims drew credibility from material dominance and corporate alignment, while China’s gained traction precisely because Washington’s rejection of multilateral norms created space for alternative scripts. Crises act as catalytic junctures, and the advent of generative AI, reframed as systemic risk since ChatGPT’s release in November 2022, combined with Trump’s sovereigntist turn, has consolidated competing visions of global order rather than convergence on shared norms.

AI governance, the article suggests, may serve as a bellwether of a post-hegemonic or multiplex world order. What is at stake is not simply the regulation of a transformative technology but the capacity of global governance to sustain any authority at all in a system fractured by sovereignty claims, geopolitical rivalry and competing visions of justice. As the United States and China advance incompatible legitimacy scripts, one grounded in sovereign voluntarism and competitive dynamism, the other in UN-centred cooperation and developmental equity, the study warns that the field exhibits pluralised and antagonistic legitimation that resists consolidation, offering a preview of how authority itself may be contested across the international system in the decade ahead.

Subject of Research: The crisis of legitimacy in global AI governance under competing US and Chinese legitimation strategies

Article Title: Reflexive legitimation conflict: trumpism and the crisis of legitimacy in global AI governance

Article References: Pegram, T. (2025). Reflexive legitimation conflict: trumpism and the crisis of legitimacy in global AI governance. Global Public Policy and Governance, 5(4), 340-352. https://doi.org/10.1007/s43508-025-00123-x

Image Credits: AI Generated

DOI: 10.1007/s43508-025-00123-x

Keywords: AI governance, legitimacy, Trumpism, global governance, China, United States, multilateralism, technological nationalism, UNESCO, open-source AI, sovereignty, post-hegemonic order

The core move in the framework is definitional, and it is more consequential than it might first appear. Prunckun defines non-crime as the absence of a specified criminal act during an identifiable episode, even though the person had a practicable opportunity to commit it and faced a relevant incentive, provocation, or occasion. An episode, in this scheme, specifies the person, the possible act, the setting, and the period under examination. This episode-level focus deliberately separates non-crime from two related concepts that criminologists have long studied: long-term abstention, in which a person never offends across repeated observations, and desistance, the process by which people with offending histories stop. One honest expense claim, the framework insists, establishes neither a lifetime of virtue nor a turning point in a criminal career. It is a single outcome that demands its own explanation.

The definitional work also draws careful boundaries around what counts as an opportunity. A practicable opportunity exists when a person has the ability and access needed to commit the specified act during the study period, and the situation presents a potential benefit, grievance, or temptation connected to the offense. Crucially, the person need not consciously consider offending at all. An employee might recognize a chance for financial gain without ever viewing dishonesty as an acceptable route to it, a distinction that echoes Per-Olof Wikström’s Situational Action Theory, which separates seeing an act as a possible choice from selecting it. Complete elimination of opportunity, such as an access restriction that renders theft impossible, is treated as a distinct and valid form of prevention but is excluded from the comparison, because the framework is interested in episodes where the offense remains genuinely possible and yet does not occur.

At the heart of the theory are five categories of restraint that may explain why the offense does not happen. Moral restraint operates when a person rejects an act because it violates an ethical standard they accept or conflicts with the kind of person they believe themselves to be, whether or not discovery is likely. Prudential restraint stems from expected costs, including legal punishment, dismissal, loss of income, or damage to future prospects; a narrower version, detection-only restraint, depends specifically on the perceived risk of being caught and the consequences expected to follow. Relational restraint arises from obligations, trust, belonging, and accountability to particular people, so that deception feels unacceptable because it would betray a colleague even if no practical loss would result. Situational restraint covers practical barriers that make an offense harder but still possible, such as extra procedural effort, guardians who might interrupt, or delays that sap the appeal of completing the act. Legitimacy-based restraint, finally, operates when a person follows a rule because they regard it as proper or believe the authority issuing it has the right to do so, a mechanism long associated with Tom Tyler’s procedural justice research.

None of these categories is entirely new, and Prunckun is explicit that the contribution lies elsewhere. The framework builds on Travis Hirschi’s social control theory, in which attachment, commitment, involvement, and belief explain conformity; on Walter Reckless’s containment theory, with its inner and outer buffers against delinquency; and on Anthony Bottoms’s distinction between prudential, normative, situational, and habitual compliance. What the new framework adds is the insistence that these processes be treated as separable, measured independently, and compared within a single episode. A social bond such as employment, the article notes, can supply several kinds of restraint at once: fear of losing a career is prudential, obligation to colleagues is relational, commitment to professional integrity is moral, and respect for the institution’s right to regulate conduct is legitimacy-based. Treating the bond as a single protective force, the argument goes, obscures which process actually did the work.

The framework’s most provocative claims are four conditional propositions about what happens when restraints change. The first holds that when one restraint weakens, the probability of non-offending should decline less among people whose conduct is also supported by other relevant restraints, and more among those who depended primarily on the weakened one. The second proposition is a detection-focused version: an equally large drop in perceived risk of being caught should produce a bigger decline in non-offending among people whose restraint depends mainly on detection, while leaving others largely unaffected, particularly where a moral filter prevents the offense from even entering consideration. The third proposition turns within-person, predicting that weakening a restraint in one setting, compared with another setting where the same person faces comparable opportunities, should reduce non-offending only when that restraint’s activation conditions are present. The fourth extends the logic to desistance, proposing that increases in separately measured restraints should predict continued non-offending among people with offending histories, with the greatest benefit appearing after a previously dominant restraint has weakened.

These propositions are framed as empirical questions rather than settled findings, and the article is candid about conflicting evidence. David Nagin’s review of deterrence research found more consistent support for the likelihood of being caught than for punishment severity as a deterrent, while studies of whether morality conditions deterrence have produced mixed results: Raymond Paternoster and Sally Simpson found sanction threats mattered most when moral inhibitions were weak in corporate crime scenarios, whereas later work found perceived certainty associated with offending even among participants with stronger moral beliefs. The framework also warns against circular reasoning. Inferring that someone depended on detection merely because they offended after monitoring was reduced would render the explanation unfalsifiable. Restraints must be measured before conduct, using indicators such as moral judgments about the specific offense, perceived risk, expected losses, obligations to particular people, beliefs about an authority’s rightness, and observable practical barriers.

A hypothetical workplace example illustrates the machinery. Imagine two employees who both submit accurate expense claims, though each could profit from a false one. Independent evidence shows the first is supported by moral objections, workplace attachment, and acceptance of the rule’s legitimacy, while the second relies mainly on the expected risk and consequences of detection. If reduced monitoring lowers both employees’ perceived risk by the same amount, the framework predicts a larger increase in the second employee’s likelihood of offending. Other changes help disentangle the processes: a breakdown in workplace relationships weakens accountability without touching expected sanctions, perceived unfairness erodes legitimacy without altering moral objections, and relaxed verification makes the false claim easier without changing either. By contrast, an access restriction that makes the false claim impossible for both employees removes the episode from the comparison entirely, because prevention by eliminating opportunity is a different phenomenon from restraint while opportunity remains.

The framework also situates individual restraint within broader social conditions, assigning them three distinct roles. Social influences may act as antecedents that help develop restraints, as contextual conditions affecting access to opportunities and resources, or as moderators that alter when and how strongly restraints operate. Research linking neighborhood collective efficacy to lower violence, for example, does not by itself identify what prevents violence in a particular episode; the framework asks whether the mechanism runs through supervision, valued obligations, or something else. Gendered patterns of access to positions of trust shape opportunities for corporate fraud, suggesting that observed sex differences may reflect roles and exposure rather than inherent differences in moral restraint. Childhood self-control, linked in a New Zealand birth cohort to later criminal conviction, may affect whether a person can follow through on a commitment when competing motives arise, while genetic associations with self-regulation traits do not by themselves establish a biological pathway to non-crime.

The practical implications cut in uncomfortable directions. Evidence that multiple restraints are present does not show that withdrawing an effective restriction would be safe, because moral commitments, relationships, and perceived legitimacy may weaken or conflict, and apparently distinct restraints may depend on the same institutional arrangement and fail together. The framework likewise cautions against labeling people as uniformly risky or safe: prior offending may improve average-risk predictions, but explaining a specific episode requires evidence about the processes operating at that moment. The author acknowledges real limitations, including overlapping categories, self-reports that may overstate principled motives, and the difficulty of establishing that a practicable opportunity existed independently of motivation and conduct. Yet the central challenge stands. Offending has causes, and so, the article argues, does its absence. When a realistic opportunity remains and the crime still does not occur, that outcome is not a null result to be ignored but a phenomenon requiring its own evidence, its own measurement, and, at last, its own theory.

Subject of Research: A criminological theory of non-crime explaining why offenses do not occur when opportunities remain available

Article Title: Why Crime Does Not Occur: Toward a Theory of Non-crime

Article References: Prunckun, H. (2026). Why Crime Does Not Occur: Toward a Theory of Non-crime. American Journal of Criminal Justice. https://doi.org/10.1007/s12103-026-09954-8

Image Credits: AI Generated

DOI: 10.1007/s12103-026-09954-8

Keywords: criminology, non-crime, restraint, social control theory, deterrence, compliance, desistance, legitimacy, situational action theory, crime prevention, self-control, criminal justice

The study, conducted by Iván Sánchez-López and Gemma San Cornelio of the Universitat Oberta de Catalunya in Barcelona together with Heleny Méndiz-Rojas of the Catholic University of the North in Chile, took a deliberately targeted approach. Rather than attempting an exhaustive census of the literature, the team mined Scopus and Web of Science for publications explicitly engaging both artificial intelligence and creativity, screened them through a PRISMA-informed procedure, de-duplicated the results down to 71 unique records, and then ranked them by citation prominence. The final corpus comprised the 40 most citation-prominent studies, spanning journal articles and full conference papers, all published between 2020 and 5 February 2025.

Once the corpus was assembled, the researchers analyzed it in ATLAS.ti using Grounded Theory and the Constant Comparative Method. Initial coding produced 227 descriptive code entries, refined into 203 first-order codes, which axial coding then organized into nine categories through 230 code-category assignments. A final narrative synthesis distilled eight cross-cutting transformation features. The team used the software’s AI-assisted coding only as a supplementary first pass; every machine suggestion was manually checked, and the final codes, categories, and interpretations were settled by the researchers themselves.

The central finding is that creativity, as portrayed across this influential literature, has become a socio-technical assemblage rather than an individual capacity. Creative outcomes now emerge from the interplay of human-model interaction, institutional gatekeeping, infrastructural asymmetries, and normative conflict. Generative AI is consistently associated with expanded productivity: faster ideation, rapid prototyping, broader exploration of possibility spaces, and new formats of human-AI collaboration. But the same studies repeatedly document a parallel intensification of trouble around authorship, originality, recognition, labour, bias, access, legitimacy, and sustainability.

The reorganization of creative work follows a recognizable pattern. As the cost of producing drafts and variants collapses, the bottleneck migrates from generation to framing, prompting, selecting, editing, and justifying outputs. In human-computer interaction and design research, this appears as iterative prompt craft, co-creative interfaces, and curation as the central creative acts. In education, generative tools are framed simultaneously as scaffolds for writing, feedback, and idea generation, and as threats to assessment, academic integrity, and critical thinking. In management and innovation studies, the emphasis falls on productivity and capability building, with experimental and empirical work showing that augmentation outcomes depend heavily on task conditions, organizational resources, and workflow integration rather than unfolding as a uniform boost to creativity.

The review argues that these expansions do not distribute their benefits evenly. Generative tools may lower the barriers to producing plausible artifacts, yet sector reviews suggest they can concentrate advantage around computing infrastructure, technical expertise, and validation authority. Barriers shift from production toward evaluation, credibility, and institutional endorsement, and access to tools does not automatically translate into recognition or effective use. The review also identifies a mismatch between productive acceleration and evaluative capacity: when systems flood the pipeline with candidates, criteria for quality, originality, and appropriateness do not scale automatically, and selection can drift toward popularity defaults, raising the risk of homogenization even as output volume rises.

Questions of evaluation are further complicated by evidence of human bias against machine-made work. Studies in the corpus position evaluators as gatekeepers whose identity-based perceptions of whether an artifact is human- or AI-produced shape recognition and uptake. In Csikszentmihalyi’s systems view of creativity, creative status is conferred by a field of evaluators, and generative AI disrupts that field function by multiplying and obscuring the sites of evaluation: merit, originality, and responsibility are now negotiated across platforms, curators, clients, reviewers, and model providers, while institutions cling to accountability logics designed for individualized authorship.

Reading the corpus against contemporary scholarship exposes two significant gaps. First, sustainability receives comparatively thin treatment. While the reviewed studies address it mainly through urban and infrastructural perspectives, recent research has developed systematic accounts of the material externalities of large-scale generative systems, including electricity demand, carbon and water footprints, data-centre infrastructure, and disclosure obligations. Second, Global South and minoritized perspectives remain weakly integrated into the dominant corpus, even though training data distributions, dominant-language resources, and uneven access to computation may act as technical and cultural priors that determine which outputs and epistemic positions become visible in the first place.

The authors are candid about the limits of their design. Focusing on the most-cited publications maps dominant framings but may reproduce existing epistemic hierarchies, since citational prominence is treated as an indicator of visibility and influence rather than quality or truth. The corpus also reflects indexed, anglophone, predominantly Global North academic infrastructures, and the 2025 slice of coverage is partial by construction. The researchers call for future work using ethnographies of everyday use, comparative studies across languages and cultures, and mixed methods connecting computational traceability with critical analysis of labour and environmental costs.

The study’s principal interpretive conclusion is that expanded productivity and contested authority are not separate outcomes but interrelated dimensions of AI-mediated creativity. Artificial intelligence does not simply automate creative tasks; it relocates where creative agency sits, changes how contribution is evaluated, and reshapes which actors and infrastructures participate in producing legitimacy. High-impact scholarship, the authors argue, does not merely describe this transformation. It also helps stabilize influential definitions of creative value and redistribute forms of recognition, making it all the more important to examine how authority, visibility, and material conditions determine which forms of AI-mediated creativity become visible and legitimate.

Subject of Research: A citation-informed review of how highly cited research frames the transformation of creativity under artificial intelligence

Article Title: Expanded productivity, contested authority: creativity and AI in high-impact research

Article References: Sánchez-López, I., Méndiz-Rojas, H., & San Cornelio, G. (2026). Expanded productivity, contested authority: creativity and AI in high-impact research. AI & SOCIETY. https://doi.org/10.1007/s00146-026-03213-4

Image Credits: AI Generated

DOI: 10.1007/s00146-026-03213-4

Keywords: artificial intelligence, creativity, generative AI, authorship, human-AI collaboration, creative industries, evaluation, legitimacy, bias, sustainability, research review, AI & Society