Aug 2026· Management Decision· pp. 1-27· 0 citations· 57 references
TL;DR
It is shown that the “dark side” of AI governance is not reducible to isolated ethical failures but emerges as an institutional outcome of competing governance logics, shifting attention from technical risks alone to the organizational architectures through which responsibility, control and legitimacy are continuously negotiated.
Abstract
This study proposes an integrative framework to explain how generative AI firms organize and adapt governance architectures to temporarily contain the conflicting logics of ethics, profit, stakeholder inclusion and technological autonomy.
Drawing on paradox theory, the study employs a multiple-case design analyzing ten frontier GenAI firms: OpenAI, Anthropic, Google DeepMind, Mistral AI, xAI, Inflection AI, Aleph Alpha, DeepSeek, Meta and Stability AI. A novel three-dimensional analytical framework operationalizes governance tensions through systematic documentary coding and comparative analysis.
The results identify five governance archetypes in GenAI firms: Contested Equilibrium, Institutionalized Ambiguity, Amoral Drift, Gilded Cage and the Founder’s Paradox. These archetypes demonstrate how attempts to incorporate ethics and accountability can create new forms of opacity, dependency and concentration of power. We interpret this recurring pattern as a form of “institutional paradoxification”, namely a theory-building construct that captures how governance mechanisms may, under certain conditions, amplify rather than resolve the tensions they are intended to contain.
This study extends paradox theory into the governance of frontier GenAI firms by showing that the “dark side” of AI governance is not reducible to isolated ethical failures but emerges as an institutional outcome of competing governance logics. In doing so, it shifts attention from technical risks alone to the organizational architectures through which responsibility, control and legitimacy are continuously negotiated.
AI governance instruments are proliferating, and so are their difficulties. Across major jurisdictions and international bodies, reform efforts built on substantially different premises encounter a recognizably similar pattern of failure. I argue that anticipatory regulatory governance rests on three
operational premises
—categorical stability, epistemic accessibility, and manageable pace—and that AI’s emergence, opacity, and velocity violate all three in compound. These premises form a distinct layer of operational preconditions, not a complete theory of governance. Reform within the existing premises reproduces the violations they produce. I call this configuration the
reform trap
: a paradigmatic lock-in at the level of operational preconditions, distinct from path dependence and policy paradigm rigidity. The pattern is convergent across five strategies in active reform—categorical regulation, process-based management, information disclosure, normative guidance, and adaptive experimentation—and persists even in the most adaptive of them. I propose three
premise-level substitutions
—outcome observability, causal attributability, and enforcement capability—each replacing a violated precondition with a weaker, design-addressable one. These differ from outcome-based and performance-based regulation, which swaps instruments within an architecture whose premises remain stable.
This paper investigates how AI narratives actively constrain the horizon of conceivable governance options. Moving beyond the established finding that narratives influence policy, we argue they fundamentally prefigure the scope of the politically thinkable by creating what we term “Policy Umwelten.” Adapting Jakob von Uexküll’s concept, a Policy Umwelt describes the perceptible and actionable landscape of interventions available within a given narrative, which systematically renders alternatives invisible. We analyze three competing narratives: Existential Risk, Accelerationism, and Critical AI. We do so through a “What’s the Problem Represented to Be?” analysis of their foundational manifestos. Each constructs a distinct Umwelt: Existential Risk makes global technocratic containment imperative; Accelerationism legitimizes only deregulation and market liberation; and Critical AI centers structural reform. Despite their overt conflict, these narratives converge in constructing Policy Umwelten that render a pragmatic approach to governance difficult to perceive: one that emphasizes iterative learning from empirical feedback and institutional flexibility over grand visions or predetermined certainties. The paper concludes that the most significant constraint on AI policy is not a lack of options but a collective failure of political perception. The Policy Umwelt concept thus provides a diagnostic tool for mapping ideological deadlock and identifying pathways toward more adaptive and less deterministic technological futures.
G. Papyshev, Ninell Oldenburg· The International Journal of...· 0 citations
The Qualitative Model of Ethics (QME), a naturalistic and teleological framework grounded in a single evaluative scalar: the generative capacity of the Whole Living System (WLS), provides a unified and operational framework for AI governance at scale.
A reframing of AI unaccountability as a constitutive property of sociotechnical systems, an extension of the four barriers to accountability, and a practical instrument for identifying accountability voids in specific AI deployments are contributed.
Long Hoang Nguyen, E. Späthe, S. Lins et al.· 0 citations
The just circular economy (JCE) has emerged as a structuring normative horizon in sustainability transitions, but its analytical value remains limited by two structural gaps: the absence of a constitutive definition integrating all justice dimensions identified in the literature, and the absence of a typology capable of distinguishing genuinely integrative configurations from partial or strategic uses of the concept. Drawing on a Weberian ideal-type theorisation approach, this article advances three interrelated contributions. It proposes a constitutive and normative definition of the JCE structured around five interdependent justice dimensions derived from three theoretical pillars: multidimensional environmental justice, intergenerational equity, and circularity justice. Four dimensions, namely distributive, procedural, and recognitive justice alongside intergenerational equity, rest on a consolidated theoretical corpus; the fifth, circularity justice, is formalised as an emerging construct that renders visible the injustices internal to circular value chains that existing tripartite frameworks cannot detect, notably the transfer of costs, risks, or injustices from one actor, one territory, or one generation to another. It further develops a configurational typology of five archetypal JCE forms, specifically the redistributive, procedural, recognitive, sustainable, and integrative forms. It finally formalises circularity justice as a fifth analytical dimension, initiating a research programme centred on the construction of intra-chain justice indicators capable of measuring the just circularity gap within global value chains. Taken together, these contributions provide an operational diagnostic framework enabling the distinction between a substantive JCE and a façade JCE, and open a structured empirical agenda for future research.
Debates on artificial intelligence (AI) governance focus on legal regulation, risk management, institutional accountability, and the division of responsibilities between the state, market, and other social actors. However, the anthropological and social assumptions underlying governance models remain underexplored, particularly in relation to the risk of erasing human dignity amid the growing technocratisation and algorithmisation of social life. The study analyses Leo XIV’s reflections on AI governance. This article adopts a conceptual and comparative approach. It is based on a close reading of the encyclical Magnifica Humanitas and a systematic comparison of the framework for AI governance articulated in the document with dominant AI governance models identified in the literature. This study reconstructs an implicit hybrid model of AI governance grounded in eight principles: (1) human dignity, (2) the common good, (3) the universal destination of goods, (4) social justice, (5) solidarity, (6) subsidiarity, (7) shared responsibility, and (8) transparency/accountability. The Tower of Babel – Nehemiah heuristic demonstrates that contemporary AI governance can be understood as a fundamental tension between two logics: one oriented toward centralisation, control, and exclusion, and another toward participation, distributed responsibility, and the protection of human dignity. As an additional contribution, the study operationalises the proposed model through the example of AI-assisted allocation of social benefits, translating its normative principles into concrete governance requirements. Leo XIV’s reflections not only align with a hybrid governance model but provide a normative corrective to both state-led and market-led models by prioritising relational human dignity and shared responsibility over centralised control. The study develops a Leo XIV-informed, normatively grounded hybrid model of AI governance that extends beyond the exercise of agency over technology to fostering a more just and inclusive social order in which technology enables human flourishing.
S. Fel, Marta Choroszewicz, Jarosław Kozak· AI and Ethics· 0 citations
A new method, called CW-Net, translates the reasoning process of an autonomous vehicle’s AI system into understandable concepts that explain its behavior.