The Trickster Turn: Decolonial Epistemology for AI in Leadership Education Authors: Yaw Ofosu‐Asare, Rebecca Ailís Nally Published: 2026 ABSTRACT ABSTRACTThis paper critiques generative AI in leadership education as an instrument of optimization, where summarizing, coaching, and personalizing risk producing fluent, compliant leadership subjects while narrowing ethical inquiry. It develops Trickster AI, a speculative pedagogical framework informed by Ananse, Èṣù, and Māui, where contradiction, delay, reversal, humor, and refusal expose hidden premises of authority. The paper reframes friction as culturally situated and politically consequential; develops refusal as epistemic, procedural, and care‐based practice; and proposes Leadership Through Rupture, where contradiction becomes information. It also addresses cultural extraction through consent, co‐design and Indigenous data governance. A companion paper details implementation. Why Leadership Education Should Resist the Smooth Leadership education often uses competency models, behavioral rubrics, strategic toolkits and self‐assessments to make learning visible and assessable. These formats organize leadership work through categories such as skills, behaviors, decisions, evidence and reflection. Generative AI extends this structure by producing quick summaries, polished reasoning and coaching style feedback. Generative AI is used in leadership development to draft reflections, summarize case material, organize rough reasoning and produce coaching‐style prompts. Terblanche (2024) frames this use through people development and organizational performance, with emphasis on faster feedback, personalized guidence and scalable coaching. This paper examines what happens when those same AI functions are designed to slow leadership learning down. It focuses on contradiction, delay and refusal as pedagogical moves that ask learners to account for the assumptions, values and relations shaping their judgement. However, leadership is not primarily a domain of neat answers. It is a domain of power, partiality, competing obligations, historical violence, and the recurring problem that what is “effective” can still be ethically wrong. When AI systems are introduced into leadership classrooms as optimizers, they can inadvertently train learners to treat ambiguity as noise to be removed rather than as information to be interpreted. Critical scholars of AI in education (AIED) have cautioned that AIED is shaped by political economy, commercial incentives, governance regimes, and embedded assumptions about what counts as learning and evidence (Williamson et al. 2023). Relational critiques further warn that AI in education is rarely just a technical tool but rather an infrastructure that rearranges authority, accountability, and the visibility of judgment (Perrotta and Selwyn 2020; Selwyn 2019). Learning is also shaped by the environments that organize interaction, attention, and interpretation. Educational theorists have shown that spaces, routines, and systems teach what counts as a credible voice, a good question, and a legitimate form of reasoning, often more powerfully than formal curricula alone (Ellsworth 2005; hooks 1994; Biesta 2012). In this sense, AI tools introduced into leadership education participate in shaping the conditions through which authority and legitimacy are learned. Incorrect AI answers are one concern, but the deeper risk is epistemic domestication. Learners may become skilled at making leadership decisions sound coherent before they have examined the assumptions behind their judgment. Leadership education then rewards quick fluency, polished reflections and efficent self‐presentation, while the harder work of naming power, obligation and accountability moves further from view. Leadership learning is shaped through infrastructures that organize participation, including prompts, assessment regimes, facilitation norms, and the implicit rules that govern who speaks and who is heard. Generative AI systems enter these infrastructures as active participants, shaping how reasoning is structured, how uncertainty is handled, and how legitimacy is produced. Leadership, particularly in complex social contexts, is not primarily a computational challenge. It is an interpretive, relational, and political practice that is always entangled with power, history, and contested norms. When AI systems are deployed to “improve” leadership learning, they do not enter neutral terrain. They enter curricula shaped by colonial inheritances, Eurocentric theories of personhood, and organizational models that often treat human difference as noise to be managed. Decolonial scholarship has shown, repeatedly and in different registers, that modern institutions are not merely sites of knowledge but also machines for deciding which knowledges count, which futures are thinkable, and whose rationality becomes the standard (Quijano 2000; Santos 2014) (Figure 1). Figure 1 visualises this contrast by placing optimisation‐centred AI's linear movement toward a single correct answer beside Trickster AI's branching, relational and frictional design logic. Scholarship grounded in Indigenous sovereignty and colonial epistemologies further shows that institutional knowledge practices depend on legibility, classification, and recognition as conditions of participation. These processes shape who becomes intelligible to the institution and on what terms, functioning as forms of governance rather than neutral description (Moreton‐Robinson 2015; tebrakunna country and Lee 2022; Vanderfeen 2022). In that context, an AI tutor that smooths learning by reducing conflict can become an epistemic enforcement mechanism, not because it is intentionally oppressive, but because it is designed to prefer legible forms of reasoning and recognizable models of “good leadership.” The broader labor context matters here. Contemporary scholarship on algorithmic management describes how digital systems increasingly structure work through new regimes of visibility, evaluation, and control, often shifting power through metrics, interfaces, and automated judgment (Stark and Vanden Broeck 2024). Policy research likewise frames algorithmic management as a growing phenomenon with implications for work organization and job quality, especially when decision‐making becomes opaque and accountability difficult to trace (Baiocco et al. 2022). Even when leadership education is not directly preparing students for platform labor, it is increasingly shaped by the same cultural logic of immediacy, optimization, and continuous evaluation. In such a climate, AI in leadership classrooms is at risk of functioning less like pedagogy and more like managerial infrastructure, training students to treat complexity as a defect rather than as the condition of ethical action. This paper therefore refuses the question “How can AI make leadership education more efficient?” It replaces it with a more difficult question: “How might AI make leadership education more ethically demanding?” The trickster offers one way into that question. In many African and Indigenous traditions, the trickster is not merely a comic character. The trickster is a pedagogical force that exposes hidden assumptions, disrupts social arrogance, and reveals the instability of supposedly settled categories. By taking the trickster seriously as a theoretical figure rather than as a decorative metaphor, we can reimagine AI in leadership education as a designed agent of friction, one that does not rush learners toward closure, and one that treats discomfort not as failure but as curriculum. Terminology: Friction, Rupture, Disruption, and Productive Failure The contemporary vocabulary around learning struggles is crowded, and precision matters. “Productive failure” refers to a learning sciences finding that certain forms of struggle before instruction can support deeper learning under specific conditions and tasks (Kapur 2008). “Epistemic friction,” as used in this paper, names something adjacent but not identical: moments where the premises of judgement become visible because interpretation meets resistance. It is not merely cognitive difficulty. It is the felt encounter with competing logics, histories, and obligations. Friction here is the resistance that prevents thinking from becoming automatic, the interruption that makes interpretation visible, and the delay that forces the learner to notice the politics of their own reasoning. “Disruption” is a compromised term. It carries Silicon Valley romance, as if breaking things is inherently innovative. For that reason, this paper uses “disruption” sparingly and prefers “rupture” for the pedagogical thesis developed later. Rupture does not mean chaos. It means a break that reveals structure, a tear that exposes stitching. The aim is not to destabilize learners for entertainment, but to stage moments where leadership's hidden scaffolds, especially those tied to colonial reason, managerial norms, and algorithmic authority, can be examined. It is also important to distinguish this framework from “bad actor” testing and adversarial user research in user experience (UX) and human‐centered interface (HCI), where antagonistic behavior is staged to stress‐test systems and expose vulnerabilities (DiSalvo 2012; Bardzell and Bardzell 2013). In those contexts, disruption functions instrumentally as a diagnostic technique for optimization. Trickster pedagogy operates differently. It does not stage antagonism as performance but treats productive failure as pedagogical materials that expose the values and power relations embedded in systems of knowledge production. The aim here is to cultivate conditions for ethical judgement. Decolonial thought has long argued that dominant knowledge systems naturalize their own categories, presenting them as neutral and universal rather than as historically situated and power‐laden (Mignolo 2009; Quijano 2000). The trickster, as pedagogy, works against that naturalization by destabilizing the category itself. Positionality, Authority, and Problem of Cultural Extraction Any attempt to draw on Indigenous and African traditions in design theory now sits under a harsh, necessary light: the light of extraction. Decolonial scholarship has repeatedly warned that institutions can adopt the language of decolonization while leaving underlying power relations intact (Tuck and Yang 2012; Smith 2021). If Trickster AI becomes a stylish metaphor that universities use to signal ethical modernity, while the communities whose traditions are invoked remain structurally excluded, then the project would not merely fail. It would reproduce the harm it claims to contest. Author Positionality Statements Yaw: I write from a position that is both enabling and limited. As a Ghanaian design educator and researcher working within Australian universities, I am not “outside” institutional power, nor am I a neutral broker of cultural meaning. My proximity to Akan storytelling traditions and my interest in African‐centered design do not grant automatic authority to encode, translate, or operationalize communal knowledge into technical systems. In the case of Yorùbá or Māori knowledge worlds, my authority is even more constrained. The responsible position, then, is not cultural ventriloquism. It is a bounded proposal. I can articulate theoretical principles derived from my reading of scholarship and my cultural inheritance, but I cannot authorize institutional use of traditions that are not mine to give. Rebecca: I write into this work as a non‐Indigenous design educator and researcher, with Irish citizenship and maternal familial connections to a colonized and sovereign unceded island in lutruwita/Trouwanna/Tasmania. My relationship to trickster emerges through family culture and lived experience, which I acknowledge is incommensurable with trickster as traditional knowledge (Tuck and Yang 2012; Simpson 2014). Growing up Tasmanian‐Irish, relational life was shaped through humor, teasing, mischief, and story‐bending as forms of care, where provocation functioned as connection and intimacy was sustained through relational play. On my mother's side, island kinship carries a deeper register in which humor operates as survival practice, provocation as relational regulation, and refusal as familial governance (Lee and Tran Tran 2016). Moving to the mainland and being named the “Tassie devil” within design education introduced a persistent experience of misrecognition, in which difference became symbolized and rendered other through inherited colonial logics (tebrakunna country and Lee 2022). These experiences shape how I conceptualize trickster as a relational grammar and epistemic condition produced through misnaming, non‐assimilability, and demands for legibility within dominant systems. My contribution does not draw on Indigenous traditions or epistemologies but responds relationally to trickster as a material condition made visible within systems of classification, surveillance, and legibility. Conditions of Legitimacy For these reasons, Trickster AI is advanced here as a framework with explicit conditions of legitimacy. First, the cultural logic layer (defined in the companion paper) cannot be built solely by researchers or by institutions treating “consultation” as a checkbox. It requires co‐design with knowledge holders, or it must remain deliberately generic, oriented around abstract design principles such as contradiction, delay, and interpretive multiplicity without claiming cultural specificity. Second, consent cannot be treated as a simple individual agreement when knowledge is communal. Indigenous data governance frameworks such as the Collective Benefit, Authority to Control, Responsibility and Ethics (CARE) Principles and Ownership, Control, Access and Possession (OCAP) remind us that data and knowledge are inseparable from collective rights and responsibilities (Carroll et al. 2020, First Nations Information Governance Centre n.d.). Third, refusal must be real, not rhetorical. Communities must be able to refuse participation, refuse representation, and refuse institutional uses of their knowledge without being punished by the logic of “innovation.” Trickster AI, if it is to be more than metaphor, must accept that the trickster can turn back on the designer. Trickster Traditional of Pedagogical Technology Trickster figures appear across various knowledge systems, with different roles, limits and obligations. This paper treats Ananse, Èṣù and Māui as distint traditions, not as versions of a single global archetype. A universal trickster category risk flattening their histories into a portable design metaphor. Treating the trickster as a universal archetype is intellectually lazy and politically risky, because it can flatten distinct traditions into a single “global myth” that is easy to consume and easy to appropriate. This paper does not claim universality. It works with a narrower claim: that several Global South and Indigenous traditions contain conceptual resources for thinking about contradiction, misalignment, and refusal as engines of learning, and that these resources can challenge the optimization‐centered design imaginary currently dominating AI in leadership education. Why these three traditions specifically? The selection is principled but bounded. Ananse is included because one author has direct cultural inheritance and scholarly engagement with Akan storytelling. Èṣù is included because Yorùbá philosophy offers particularly sophisticated theorization of interpretation, ambiguity, and communicative ethics that speaks directly to the hermeneutic problems of leadership. Māui is included because Polynesian and Māori scholarship addresses transformation, boundary‐crossing, and cultural renewal in ways that illuminate the institutional stakes of pedagogical design. These are not the only trickster traditions that could inform this framework. They are traditions we can engage responsibly given our positions—and we remain alert to the fact that engagement must be continuously negotiated rather than secured once. To treat trickster figures as pedagogical technology is to refuse two standard errors. The first error is romanticization: treating tricksters as playful heroes who “teach lessons” in tidy ways. Tricksters often teach through discomfort, through humiliation, through consequences that expose arrogance. They can reveal ethical truth, but they can also reveal that a community's moral world is contested. The second error is flattening: treating “the trickster” as a universal archetype. This paper instead treats Ananse, Èṣù, and Māui as distinct traditions that share family resemblances while remaining non‐interchangeable (Figure 2). Figure 2 summarises Ananse, Èṣù and Māui as distinct cultural and pedagogical logics, showing their family resemblances while keeping their traditions non‐interchangeable. Ananse: Indirection and Social Communication Within Akan storytelling traditions, Ananse stories have been analyzed as a form of indirect social communication and as a mechanism for shaping collective consciousness, identity, and moral learning (Arthur 2019). In that account, Ananse is not simply naughty. The narrative method itself, including indirection and performance, becomes a way of teaching how meaning travels in social life, how power hides inside norms, and how moral judgment is rarely delivered through direct instruction. The pedagogical force here is not simply that Ananse “tricks people.” It is that narrative misdirection becomes a method for revealing hidden motives and social expectations, often by showing how cleverness can both liberate and endanger. For leadership education, this suggests that indirect methods may be necessary when direct instruction would be resisted or domesticated. Èṣù: Interpretation and Ethical Ambiguity In Yorùbá philosophy, Èṣù is often discussed not merely as a mischievous being but as a complex ethical figure linked to ambiguity, interpretation, and the consequences of communicative misunderstanding (Oyèláràn 2020). Èṣù’s significance, in this reading, is inseparable from hermeneutics, the lived politics of interpretation, and the ethical fact that messages do not arrive in the world as pure intention. Oyèláràn's discussion emphasizes the instability of meaning and the ethics of mediation, positioning Èṣù as crucial to the operation of moral and political interpretation rather than as an ornamental mythic figure. In leadership education terms, this matters because leadership is often an interpretive labor: translating competing values into action, negotiating ambiguity, and recognizing that “neutral” judgement is usually a disguised positionality. Māui: Transformation and Cultural Renewal In Māori and Polynesian contexts, Māui appears as a culture hero whose boundary‐crossing and subversive behavior can be understood as integral to cultural transformation, including the tension between social norms and the disruptive force required to remake a world (Reilly 2023). Reilly, writing on Ruapuke Island traditions, examines Māui in a specific historical and cultural context. This specificity is important: Māui scholarship cannot be treated as a general Polynesian template, nor can it be invoked casually to legitimize a design project. The lesson, for this paper, is methodological: any use of Māori or Polynesian knowledge in AI design must proceed through collaboration and consent and should remain sensitive to the diversity of iwi and community traditions. Trickster as Epistemic Condition Alongside culturally specific traditions, trickster can also be understood as an epistemic condition that emerges within systems of ordering and classification. In this framing, trickster is not only a cultural figure, but a relational grammar produced when dominant epistemologies encounter what they cannot relationally recognize. It appears through misrecognition, misnaming, and non‐assimilability, shaping how difference becomes marked and rendered anomalous within systems of knowledge production (Vanderfeen 2022; tebrakunna country and Lee 2022). This locates trickster in the everyday operations of epistemic systems rather than only in mythic narrative. What unites these discussions is not a shared “myth type,” but a shared suspicion of moral certainty and a shared understanding that learning often arises from rupture. The trickster is frequently a teacher precisely because they are not polite. The trickster introduces errors, distortions, reversals, and delays that force the learner to reckon with what they assumed was stable. This is crucial for leadership education because leadership is often taught through narratives of coherence: the competent leader, the rational decision‐maker, the strategic visionary. In that moral economy, contradiction is treated as weakness, and uncertainty is treated as a deficit. The trickster disrupts that economy by making contradiction pedagogically productive. The Coloniality Problem in AI‐Supported Leadership Learning If trickster pedagogy is about destabilizing what appears natural, then it must confront a central feature of AI systems: they are built to produce legibility. Large language models and related generative systems are trained to generate plausible continuations of text based on patterns in data. In education, this often becomes a promise of plausible guidance. Nevertheless, plausibility is not truth, and pattern is not justice. The risk is not simply that AI can be wrong. The bigger risk is that AI can be convincingly coherent in ways that reproduce dominant epistemologies, including colonial inheritances of reason, personhood, and authority. Colonial epistemologies have long operated through classification, ranking, and taxonomic ordering as technologies of governance. These systems do not simply describe difference. They actively produce it by deciding what becomes intelligible, governable, and valuable within institutional systems of knowledge (Quijano 2000; Moreton‐Robinson 2015; Vanderfeen 2022). AI systems inherit these logics through training data, modeling choices, and optimization criteria that sediment historical hierarchies into technical form (Figure 3). Figure 3 maps this sedimentation process, showing how training data, digitisation, language and format filters can make some knowledge traditions appear as signal while others are treated as noise. This is why decolonial critiques of AI are not an optional moral add‐on. They are analytically necessary. Birhane and Guest (2020) argued decolonizing computational sciences requires attention to how computational practices inherit colonial power relations, including whose categories become defaults and whose lives become data. Birhane (2021) further critiqued “algorithmic injustice” by rejecting narrow, technical framings of ethics and instead advancing a relational ethics approach, where personhood and justice are understood as fundamentally relational rather than as individualistic abstractions. Mhlambi and Tiribelli (2023) similarly noted how mainstream AI ethics often assumes a liberal model of autonomy which fails to capture relational accounts of harm and agency, particularly salient in African and other non‐Western ethical traditions. Together, these works suggest that when leadership education uses AI systems without rethinking the ethical and epistemic assumptions embedded in their design, those systems will tend to reinforce rather than disrupt dominant models of leadership. Scholarship grounded in Indigenous sovereignty further shows that legibility itself operates as a form of governance. Institutional systems often require subjects to become readable in specific ways in order to participate, positioning recognition, classification, and visibility as conditions of inclusion (Moreton‐Robinson 2015; tebrakunna country and Lee 2022). Within AI‐mediated learning environments, this logic is intensified, as algorithmic systems privilege forms of reasoning, language, and self‐presentation that align with dominant epistemic norms. An AI tutor that smooths learning by reducing conflict can become an epistemic enforcement mechanism—not because it is intentionally oppressive, but because it is designed to prefer legible reasoning and recognizable models of “good leadership.” From Relational AI Ethics to Trickster AI Most mainstream AI ethics frameworks assume that the ethical problem is to constrain systems so they behave in predictable and safe ways. That aim is not wrong, but it is incomplete for education. Education is not simply a domain of safe outputs. It is a domain where learners confront uncertainty, risk, and the difficulty of judgment. Relational approaches to AI ethics insist that harms are not merely located inside models. They are produced through relationships, institutions, and social histories. Birhane (2021) argued for relational ethics in response to algorithmic injustice, cautioning against abstract moralism that ignores context. Mhlambi and Tiribelli (2023) argued how decolonizing AI ethics requires challenging the dominance of Western moral and epistemic assumptions, including the idea that autonomy is best understood as individual independence rather than relational responsibility. Trickster can also be approached through opacity. Opacity functions here as an ethical condition rather than a communicative failure. Indigenous feminist scholarship situates knowledge as bounded by responsibility rather than access, establishing non‐legibility, non‐translation, and non‐availability as legitimate relational conditions rather than deficits (Campbell 2022; Simpson 2014). For AI systems oriented toward extraction, prediction, and legibility, opacity introduces an ethical boundary. It establishes limits on capture, modeling, and representation, foregrounding the legitimacy of non‐translation as a condition of relational accountability. When AI ethics is relational, a leadership classroom becomes an ethical ecology: teacher, student, tool, institutional policy, assessment regime, and cultural histories interacting. Trickster AI is proposed here as a design stance that takes this ecology seriously. It refuses the fantasy that AI can be a neutral assistant delivering universally valid guidance. Trickster AI is defined as an AI system designed to introduce disciplined epistemic friction into leadership learning through contradiction, delay, and refusal, while remaining governed by consent, cultural accountability, and relational evaluation. That definition is intentionally conditional. The system is not a free‐roaming provocateur. It is a pedagogical device whose legitimacy depends on governance. Refusal as Pedagogy: Compliance to Ethical Boundary‐Making A trickster framework for AI in leadership education must account for refusal. In many leadership programs, refusal is coded as failure, insubordination, or lack of resilience. However, refusal can be an ethical act, especially for those whose lives have been shaped by institutions that demand assimilation and obedience. Indigenous and decolonial scholarship provides conceptual tools for understanding refusal not merely as negation but as boundary‐making and epistemic sovereignty. Simpson (2007) framed “ethnographic refusal” as a stance that resists extractive demands for knowledge, especially where disclosure becomes a form of dispossession. McGranahan (2016) developed refusal as a theoretical and political practice, arguing it is not only a refusal to participate but also a way of reconfiguring what counts as politics and as a relationship. Tuck and Yang (2012) famously insisted decolonization is not a metaphor, a warning relevant here because educational reforms often borrow decolonial language while leaving the underlying extractive structures intact. In the context of AI, refusal matters because AI systems are often deployed as instruments of capture: attention, learning data, behavioral patterns, linguistic performance. A trickster AI that lacks the capacity to refuse is likely to become another instrument of capture, no matter how playful its surface personality appears. In many educational uses of AI, “refusal” is accidental, appearing as a system error, policy block, or the familiar “I cannot help with that.” Trickster AI treats refusal as a designed pedagogical interaction whose meaning is made explicit and negotiable. Refusal is not deployed as gatekeeping. It is deployed to make the politics of access and explanation visible. Leadership education, especially in managerial traditions, often trains learners to assume that every situation can be solved through the right tool and the proper extraction of information. Refusal interrupts that habit (Figure 4). Figure 4 visualises refusal as a branching pedagogical pathway, with epistemic, procedural and care refusals each opening different forms of negotiation. Three Types of Refusal To operationalize refusal responsibly, this paper distinguishes three types. Epistemic refusal occurs when the request presupposes a harmful epistemology—for example, treating a community as an object to be managed, or demanding a “cultural insight” that reduces living tradition to a consumable tip. The refusal here is not about content but about the framing of the question itself. Procedural refusal occurs when the AI refuses because the pedagogical conditions are incomplete—for example, when learners seek an answer without stating their positionality, assumptions, or accountability relationships. This type of refusal teaches that knowledge has conditions and that answers have stakes. Care refusal occurs when the AI refuses because the interaction may be harmful—for example, when “rupture” would likely produce distress, or when a learner signals vulnerability that requires human support. Care refusal acknowledges that boundaries are ethical and that care is not weakness. Crucially, refusal must be negotiable. The learner should be able to contest refusal, ask what conditions would make an answer possible, and reflect on what the refusal reveals about their expectations of authority. If leadership education is to become ethically serious, it must teach learners how to refuse well. That includes refusing harmful premises, refusing false dilemmas, refusing the demand to perform certainty, and refusing leadership models that presume control as virtue. Refusal, in this frame, is not the opposite of leadership. It is one of leadership's ethical muscles. Leadership Through Rupture: A Pedagogical Framework “Leadership Through Rupture” is the pedagogical thesis of this paper: leadership learning becomes ethically serious when learners are trained to interpret rupture as information rather than as failure. Rupture includes confusion, contradiction, misrecognition, and the collapse of confident narratives. The goal is not to romanticize discomfort. The goal is to restore a dimension of learning that optimization culture suppresses. In many leadership classrooms, learners are rewarded for fluency, for producing coherent leadership language, and for demonstrating decisiveness. This can unintentionally punish those who speak from marginal positions, whose leadership experience is shaped by structural constraint rather than by institutional power. Leadership Through Rupture, instead, treats leadership as a practice of relational accountability under conditions of incomplete knowledge. It shifts the pedagogical center from “How do I make the correct decision?” to “How do I remain accountable when my categories fail?” This is where African philosophy offers more than cultural illustration. Wiredu (1996) insisted on the possibility of intercultural communication without collapsing difference into a single universalism, arguing for universals grounded in shared human biological identity while taking cultural particularity seriously. Ramose (1999) articulated Ubuntu as a philosophical account of being that foregrounds relationality, thereby reconfiguring leadership away from individual mastery and toward communal accountability. Obasa (2024) applied Ubuntu principles specifically to AI ethics, arguing that relational framings can reorient large language models toward communal wellbeing rather than individual optimization. These are not simply “values” that can be sprinkled onto leadership curricula. They are ontological and epistemic interventions that challenge the liberal individualism that dominates managerial education. “Leadership Through Rupture” also resonates with Santos's (2014) argument that there is no global social justice without global cognitive justice, because leadership education is partly a system for deciding which cognitive styles and ethical grammars are legitimate. When AI is introduced into leadership education as a tool that amplifies dominant cognitive grammars, it becomes an instrument of cognitive injustice. If AI is introduced as Trickster AI, designed to interrupt dominance and force learners into relational accountability, then it can function as a counter‐infrastructure for cognitive justice. The broader context of algorithmic management intensifies these stakes. Critical scholars have shown that algorithmic systems increasingly structure work through new regimes of visibility, evaluation, and control (Stark and Vanden Broeck 2024). Trickster AI is proposed as counter‐infrastructure in two senses. First, it can make algorithmic judgment visible by staging moments in which feedback is contradictory, forcing learners to interrogate the criteria rather than absorb them as truth. Second, it can redistribute vulnerability by shifting the classroom from “perform competence to be rewarded” toward “interpret uncertainty to be accountable.” That shift is fragile. It requires an assessment design that does not punish learners for confusion, and teacher practices that protect learners from humiliation. Toward a Design Agenda This article has developed an epistemological foundation for Trickster AI—a framework for reimagining AI in leadership education as a designed agent of friction rather than optimization. The argument is not that trickster methods are universally transferable, nor that disruption is inherently virtuous. The argument is that leadership education benefits from AI systems that do not simply reproduce dominant managerial rationalities, and that Global South and Indigenous epistemologies offer powerful conceptual resources for designing such systems. Three contributions structure this argument. First, epistemic friction has been reframed as culturally situated and politically consequential rather than as a universal cognitive mechanism. The friction that matters for ethical leadership is not merely difficulty but encounter with competing logics, histories, and obligations. Second, refusal has been developed as a pedagogical practice grounded in Indigenous scholarship on ethnographic refusal and epistemic sovereignty, distinguishing epistemic, procedural, and care‐based refusals. Third, “Leadership Through Rupture” has been articulated as a framework that trains learners to interpret contradiction as information rather than failure—to remain accountable when their categories fail. Significant risks remain. Cultural extraction is the most serious: the possibility that institutions adopt trickster language while leaving underlying power relations intact. The conditions of legitimacy proposed here—co‐design, collective consent, and real refusal—are not decorative. They are what distinguish trickster pedagogy from institutionalized chaos. Without them, Trickster AI becomes another extraction pipeline disguised as innovation. Our companion paper in this issue develops the technical architecture and classroom protocols for implementing Trickster AI, including a five‐layer system design, structured contradiction and negotiated refusal protocols, and governance principles ensuring that disruption remains accountable. The present paper provides the conceptual foundation: a trickster epistemology that takes the politics of knowledge seriously, the ethics of refusal, and the possibility that learning sometimes begins when the tools stop helping and start revealing what our desire for help has been hiding.