CoChat
Sign Up
Cited by 7Indigenous Health, Education, and RightsEthics and Social Impacts of AIDigital Education and Society

Towards a decolonial AI sovereignty model for indigenous educational settings

Open access
Mbah, Marcellus Forh, Jarvis, Leigh·2026·Cogent Education
DOI 10.1080/2331186x.2026.2626631
Open this paper in CoChat — free accountOpen PDF

Abstract

This article proposes a critical reorientation of Generative Artificial Intelligence (GenAI) in Indigenous educational settings. We argue that GenAI, built on Western epistemologies, reproduces colonial hierarchies and risks what de Sousa Santos calls epistemicide. To confront these harms, we introduce the Decolonial AI Sovereignty (DAIS) model – a conceptual framework grounded in Indigenous data sovereignty, relational ethics, and decolonial refusal. To operationalise DAIS, we draw on seven illustrative case studies of Indigenous-led or Indigenous-engaged AI initiatives. Together, these cases demonstrate how Indigenous communities can actively shape technological futures that challenge extractive models and assert sovereign control over knowledge. We argue that Indigenous governance of GenAI must extend beyond data inclusion to encompass full epistemic jurisdiction, including the right to withhold participation in extractive digital systems. Unlike posthuman or critical pedagogical frameworks that emphasise AI as a co-creator, DAIS is unique in grounding AI’s role within Indigenous sovereignty, refusal, and ceremonial protocols, placing political and colonial conditions at the foundation rather than the margins. By bridging critical Indigenous theory, AI ethics, and educational futures, this article contributes to an emerging discourse on Indigenous AI sovereignty and offers a roadmap for reclaiming generative technologies as tools of resurgence, not assimilation.

AI snapshot

Written from the paper’s full text.

TL;DR

The article proposes a Decolonial AI Sovereignty (DAIS) model for Generative AI in Indigenous educational settings, arguing that GenAI reproduces colonial epistemic hierarchies and risks epistemicide. Drawing on seven illustrative Indigenous-led or Indigenous-engaged AI cases, it calls for Indigenous epistemic jurisdiction, refusal, and governance beyond inclusion or bias correction.

Key findings

  • GenAI is described as structurally misaligned with Indigenous knowledge systems—oral, land-based, intergenerational, and relational—and as extending data colonialism and epistemic erasure rather than merely exhibiting technical bias.
  • Across seven cases, the authors identify recurring patterns: structural dependency when engagement lacks infrastructural control; refusal as critique and pedagogy; co-creation under constraint that rarely alters systemic architectures; and stronger sovereignty where Indigenous authority is structurally embedded.
  • The DAIS framework centres “GenAI as Ceremony” and five interlinked pillars: Epistemic Sovereignty as a cross-cutting foundation, Indigenous Legal Orders, Reverse Tutelage, Co-Creation in Education, and Relational Protocols.
  • The paper argues that ethical AI governance must move from inclusion and dataset diversification to enforceable Indigenous jurisdiction over datasets, models, deployment, and the right to withhold participation.

Methods

  • Secondary qualitative multiple-case study design with structured cross-case synthesis, using seven cases that met inclusion criteria requiring Indigenous-led governance or substantive Indigenous participation, engagement with GenAI or adjacent AI, educational/epistemic context, and sufficient empirical detail.
  • Adaptive, iterative search in Scopus, Web of Science, Google Scholar, and targeted web sources, supplemented by backward citation tracking; search terms combined Indigenous identity, AI technologies, and education, continuing until conceptual saturation.
  • Analysis proceeded in two stages: within-case descriptive and analytic memos, then cross-case synthesis through comparative matrices examining leadership, technology type, governance, and pedagogical application; themes were refined through investigator triangulation.

Limitations

  • The paper states that empirical work is scarce and sources are heterogeneous, and that standard systematic or scoping review protocols were not used because they would have excluded key sources.
  • The paper states that one case, “Rooted in the Collective,” did not include Indigenous participants and was retained as an analytic counterpoint; it also notes the work is based on secondary sources and required no institutional ethical approval.
  • Evident from the text: DAIS is an inductively developed conceptual framework from seven diverse cases, not an empirically validated or generalisable model, and the cases vary widely in geography, governance, and technology, limiting claims about causal outcomes.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1080/2331186x.2026.2626631

Cites

49 of 49 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

41 of 86 cited sources checked · 4 supported · 2 flagged · accuracy not yet measured

  • Findingreasoning

    In the Abundant Intelligences initiative, refusal, not reform, is posed as a necessary stance.

    “In this context, refusal, not reform, is posed as a necessary stance.”
    Quote in the paperSubstantial claim1 of 2 sources checked
    • cites Lewis et al., 2024

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites Silano, 2024 source

      The Introduction’s case study describes the Abundant Intelligences project as integrating Indigenous knowledge into mainstream AI through community engagement and co-creation. The Conclusion likewise advocates incorporating Indigenous perspectives into AI development and policy. These passages propose changing how AI is developed, rather than posing refusal instead of reform as necessary; page numbers were not available in our copy. (LLM, conf=0.83)

      Checked by a reasoning model — may be wrong.

      Experimental — accuracy not yet measured
  • Findingcitation

    In the U.S. tribal participatory design project, the process relied on pre-trained, non-Indigenous platforms whose datasets, design logics, and governance remained outside community control.

    “Although framed as co-creation, the process relied on pre-trained, non-Indigenous platforms whose datasets, design logics, and governance remained outside community control (Wang et al., 2024).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Wang et al., 2024 source

      The Interview Processes passage describes U.S. tribal members reviewing an existing generative-AI chatbot’s responses, and the Discussion says those responses lacked tribal input and reflected non-tribal perspectives. The paper does not describe a participatory platform-design project or establish that the platform’s training datasets, design logic, and governance remained outside community control. (LLM, conf=0.84)

      Experimental — accuracy not yet measured
  • Interpretationcitation

    GenAI outputs embed linguistic, epistemic, and cultural assumptions that frequently conflict with Indigenous worldviews.

    “Consequently, their outputs embed linguistic, epistemic, and cultural assumptions that frequently conflict with Indigenous worldviews (Kay et al., 2024).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Kay et al., 2024 source

      The Generative Hermeneutical Ignorance section says generative AI can misrepresent marginalized groups because it lacks contextual and cultural understanding. The Hermeneutical Access Injustice section also describes English-centric models that serve underrepresented languages unevenly. Our copy does not establish that outputs frequently conflict with Indigenous worldviews specifically; its mentions of Indigenous peoples appear in the references. (LLM, conf=0.86)

      Experimental — accuracy not yet measured
  • Premisecitation

    Communities from which Indigenous knowledge originates are often structurally excluded from decision-making, ownership, and benefit-sharing.

    “Yet the communities from which this knowledge originates are often structurally excluded from decision-making, ownership, and benefit-sharing (Nemorin, 2024).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Nemorin, 2024 source

      Nemorin argues that Indigenous perspectives are absent from the AI-in-education frameworks she examines and that local knowledges are not incorporated into ethics-building (pp. 1266–1267). This supports the decision-making aspect in that setting, but the paper does not establish the broader claim that originating communities are often excluded from ownership or benefit-sharing. (LLM, conf=0.86)

      Experimental — accuracy not yet measured
  • Premisecitation

    GenAI systems and related AI systems operate on vast datasets dominated by Western languages, logics, and epistemologies.

    “Both operate on vast datasets dominated by Western languages, logics, and epistemologies (Mohamed et al., 2020).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Mohamed et al., 2020 source

      Section 1 describes large amounts of data as a factor in AI advances, and §4.2 says researchers’ choices determine which knowledge enters a dataset. The paper also discusses colonial power in data practices (§3), but our copy does not establish that AI datasets are dominated by Western languages, logics, and epistemologies, or make that claim specifically about GenAI. (LLM, conf=0.84)

      Experimental — accuracy not yet measured
  • May combine two claims. 33
  • Classified as probably not a substantial claim. 23

  • May combine two claims.

    Findingcitation

    IBM Brazil's collaboration on Guarani Mbya and Nheengatu languages included consent protocols but kept ultimate control-ownership, deployment, and technical design-within institutional hands, privileging text-centric norms over oral traditions.

    “IBM Brazil's collaboration on Guarani Mbya and Nheengatu languages included consent protocols but kept ultimate controlownership, deployment, and technical designwithin institutional hands, privileging text-centric norms over oral traditions (Pinhanez et al., 2024).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Pinhanez et al., 2024 source

      Split into 2 atoms — 0 confirmed, 1 partially supported, 1 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

      Checked by a reasoning model — may be wrong.

      Experimental — accuracy not yet measured

  • May combine two claims.

    Interpretationcitation

    GenAI and structurally related AI systems fail to recognise or respect Indigenous modes of knowing, reasserting Western frameworks of knowledge production and reinforcing historical dynamics of classification, devaluation, and erasure.

    “As a result, GenAI and structurally related AI systems fail to recognise or respect Indigenous modes of knowing, reasserting Western frameworks of knowledge production and reinforcing historical dynamics of classification, devaluation, and erasure (Kay et al., 2024;Mohamed et al., 2020).”
    Quote in the paperSubstantial claim2 of 2 sources checked
    • cites Kay et al., 2024 source

      The Generative Hermeneutical Ignorance section (page unavailable) says generative AI can misrepresent or erase marginalized groups because it lacks contextual and cultural understanding, and can privilege dominant narratives. It does not specifically establish a failure to respect Indigenous modes of knowing or a reassertion of Western knowledge-production frameworks. The paper distinguishes generative AI from classification algorithms rather than treating their harms as the same historical dynamic. (LLM, conf=0.86)

      Experimental — accuracy not yet measured
    • cites Mohamed et al., 2020 source

      Section 3, “Algorithmic Coloniality” (page unavailable), says AI can carry forward inherited knowledge systems and colonial power structures. Section 4.2 says AI researchers have power to decide what counts as valid knowledge and what datasets omit. The paper does not specifically assess GenAI or establish that AI systems fail to respect Indigenous modes of knowing through classification, devaluation, and erasure. (LLM, conf=0.87)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Findingcitation

    In the Brazilian teacher training programme, no models were retrained and no Indigenous datasets were incorporated, leaving participants dependent on default, externally owned infrastructures.

    “Despite this positioning, no models were retrained and no Indigenous datasets were incorporated, leaving participants dependent on default, externally owned infrastructures (Fausto et al., 2024).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Fausto et al., 2024 source

      The section “Integração da Inteligência Artificial no Curso” (page not provided) says teachers used DALL-E, Midjourney, Leonardo and Copilot to create culturally relevant images. It does not say whether models were retrained or Indigenous datasets incorporated, nor does it establish that participants remained dependent on externally owned infrastructure. (LLM, conf=0.82)

      Experimental — accuracy not yet measured

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationcitation

    GenAI must be codesigned from the ground up, respecting Indigenous ontologies of relationality, land-based ethics, kinship, and intergenerational learning.

    “Instead, GenAI must be codesigned (Leal Filho et al., 2024) from the ground up, respecting Indigenous ontologies of relationality, land-based ethics, kinship, and intergenerational learning.”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites Leal Filho et al., 2024 source

      Table 7 (Additional information; page not provided) describes a participatory co-design approach supporting Indigenous values in rural Indonesia. That supports the general co-design idea, but the cited paper does not establish that GenAI must be co-designed or specify relationality, land-based ethics, kinship, and intergenerational learning as design requirements. (LLM, conf=0.83)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Findingcitation

    Kaiapuni educators described LLMs as epistemically unreliable-producing incorrect, unsafe, or culturally inauthentic outputs, omitting Indigenous knowledge, and risking algorithmic commodification of cultural heritage, threatening language revitalisation work.

    “Educators described LLMs as epistemically unreliableproducing incorrect, unsafe, or culturally inauthentic outputs, omitting Indigenous knowledge, and risking algorithmic commodification of cultural heritage, threatening language revitalisation work (Mhasakar et al., 2025).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Mhasakar et al., 2025 source

      Split into 2 atoms — 1 confirmed, 1 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

      Experimental — accuracy not yet measured

  • May combine two claims.

    Findingcitation

    Participants in the U.S. tribal project identified stereotyped, inaccurate, and missing representations, demonstrating that even the promise of inclusion can be hollow without control over the underlying data infrastructure.

    “Participants identified stereotyped, inaccurate, and missing representations, demonstrating what Couldry and Mejias argue: even the promise of inclusion can be hollow without control over the underlying data infrastructure (2019).”
    Quote in the paperSubstantial claim1 of 2 sources checked
    • cites Wang et al., 2024 source
    • cites Couldry & Mejias, 2019

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Findingcitation

    In classrooms, GenAI results in tangible pedagogical harms: automated grading calibrated to Euro-Western norms, misclassification of Indigenous words as 'errors', and visual systems that default to Western ecological framings.

    “In classrooms, this results in tangible pedagogical harms: automated grading calibrated to Euro-Western norms, misclassification of Indigenous words as 'errors', and visual systems that default to Western ecological framings (Fausto et al., 2024).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Fausto et al., 2024 source

      The Methodology and Integration of AI in the Course sections describe Indigenous teachers using generative AI to make culturally relevant images. The Discussion reports positive educational uses, but does not assess automated grading, Indigenous words being marked as errors, or whether visual systems default to Western ecological framings. Searches for those harms and for cultural bias found no passage addressing them. (LLM, conf=0.73)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Interpretationcitation

    Open-source ecosystems such as Hugging Face and BLOOM reduce technical barriers but can reproduce extractive defaults without Indigenous authority over datasets, objectives, and deployment pathways.

    “Open-source ecosystems such as Hugging Face and BLOOM reduce technical barriers but can reproduce extractive defaults without Indigenous authority over datasets, objectives, and deployment pathways (BigScience Workshop, 2022;Wolf et al., 2020).”
    Quote in the paperSubstantial claim1 of 2 sources checked
    • cites BigScience Workshop, 2022

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites Wolf et al., 2020 source

      Section 4, “Community Model Hub” (page unavailable), says Hugging Face’s hub makes pretrained models easy to access and use, supporting the reduced-barriers portion. It does not address BLOOM or the claim that open-source ecosystems can reproduce extractive defaults without Indigenous authority over datasets, objectives, and deployment. (LLM, conf=0.87)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Findingcitation

    In Brazil, educators adapted tools like DALL·E and Copilot for ecological storytelling, but these efforts were confined to prompt-level tweaks, with no authority over datasets or model logic.

    “In Brazil, educators adapted tools like DALLÁE and Copilot for ecological storytelling, but these efforts were confined to prompt-level tweaks, with no authority over datasets or model logic (Fausto et al., 2024).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Fausto et al., 2024 source

      The Methodology and ‘Integration of Artificial Intelligence in the Course’ sections describe Indigenous teachers in Brazil using DALL-E and Copilot to create culturally relevant illustrations and other visual materials. The Results section includes imagery of a river community and its natural environment, but the paper does not describe the activity as ecological storytelling. It also does not establish that the teachers’ efforts were confined to prompt-level tweaks or assess their authority over datasets or model logic. (LLM, conf=0.86)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Interpretationcitation

    Refusal is not withdrawal; it is an affirmative method to safeguard language, resist extractive infrastructures, and insist that infrastructural choices answer to Indigenous legal orders.

    “Refusal here is not withdrawal; it is an affirmative method to safeguard language, resist extractive infrastructures, and insist that infrastructural choices answer to Indigenous legal orders (Hudson et al., 2023).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Hudson et al., 2023 source

      The 'Right to refuse' section describes an active, ongoing right to reject uses of Indigenous data, and 'Right to possess' places data under Indigenous legal control rather than exploitative institutional practices. The Discussion says Indigenous rights should inform evolving data infrastructures. The paper does not specifically describe refusal as a way to safeguard language, and it explicitly includes withdrawal of previously granted permission as one possible exercise of refusal. (LLM, conf=0.84)

      Experimental — accuracy not yet measured

  • Classified as probably not a substantial claim.

    Interpretationcitation

    Reverse Tutelage reconfigures design authority by requiring AI institutions to learn from Indigenous epistemologies rather than incorporating them symbolically.

    “Reverse Tutelage: reconfigures design authority by requiring AI institutions to learn from Indigenous epistemologies rather than incorporating them symbolically (Mohamed et al., 2020;Wilson, 2008).”
    Quote in the paperSubstantial claim1 of 2 sources checked
    • cites Mohamed et al., 2020 source

      Section 4.2 describes reverse tutelage as reciprocal learning between centres and peripheries, and calls for meaningful intercultural dialogue rather than paternalistic practice. The Design section connects it to community-engaged AI design. These passages support a shift in whose knowledge shapes design, but do not specifically require learning from Indigenous epistemologies or explicitly contrast that with symbolic incorporation. Page numbers were not available in our copy. (LLM, conf=0.84)

      Experimental — accuracy not yet measured
    • cites Wilson, 2008

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Classified as probably not a substantial claim.

    Interpretationcitation

    Epistemic Sovereignty requires governance boards, binding agreements, and technical protocols granting communities authority over datasets, models, and deployment.

    “It requires governance boards, binding agreements, and technical protocols granting communities authority over datasets, models, and deployment (Christen & Anderson, 2019b;Hudson et al., 2023).”
    Quote in the paperSubstantial claim1 of 2 sources checked
    • cites Christen & Anderson, 2019b

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites Hudson et al., 2023 source

      The “Right to govern” section (page unavailable) supports community authority over data and participation in developing protocols. The “Right to possess” section (page unavailable) describes Data Governance Agreements as a means of bringing data under Indigenous governance. The cited paper does not establish governance boards or binding agreements as requirements, or specifically extend that authority to models and deployment. (LLM, conf=0.87)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Study design and conductcitation

    The study employed an adaptive, iterative search strategy appropriate for emergent fields because standard systematic or scoping review protocols assume uniform evidence bases and would have excluded key sources.

    “Because standard systematic or scoping review protocols assume uniform evidence bases and would have excluded key sources, we employed an adaptive, iterative search strategy appropriate for emergent fields (Arksey & O'Malley, 2005;Levac et al., 2010).”
    Quote in the paperSubstantial claim2 of 2 sources checked
    • cites Arksey & O'Malley, 2005 source

      The Methodological framework supports an adaptive, iterative search that can revise terms and include literature regardless of study design. But it describes this as a feature of scoping studies, not a departure from standard scoping protocols. The paper does not say those protocols assume uniform evidence bases or would have excluded the citing study’s key sources. (LLM, conf=0.87)

      Experimental — accuracy not yet measured
    • cites Levac et al., 2010 source

      The Background describes scoping studies as suitable for fields with emerging evidence because they can include varied study designs and grey literature. In Discussion, under “Framework stage three: Study selection,” the authors recommend refining the search strategy as abstracts are retrieved and describe study selection as iterative. The paper does not say that standard scoping protocols assume uniform evidence bases or would have excluded key sources; its description of scoping studies as inclusive cuts against that rationale. (LLM, conf=0.86)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Findingcitation

    Brazilian educators localised outputs for ecological narratives; U.S. tribal co-design sessions exposed stereotyped imagery in generative models but lacked authority to correct underlying systems.

    “Brazilian educators localised outputs for ecological narratives; U.S. tribal co-design sessions exposed stereotyped imagery in generative models but lacked authority to correct underlying systems (Wang et al., 2024).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Wang et al., 2024 source

      The Preliminary Findings (Interview Processes; page unavailable) describe U.S. tribal members reviewing chatbot responses in interviews and identifying biased or missing information. The Discussion describes colonized perspectives and disrespectful tribal names. Our copy does not describe Brazilian educators, ecological narratives, stereotyped imagery, co-design sessions, or participants lacking authority to change the underlying systems. (LLM, conf=0.88)

      Experimental — accuracy not yet measured

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

    Findingnone given

    The seven cases selected reflect diverse geographies and technological orientations, illustrating how generative and adjacent AI systems intersect with Indigenous governance and education.

    “The seven cases selected reflect diverse geographies and technological orientations, illustrating how generative and adjacent AI systems intersect with Indigenous governance and education.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingcitation

    In the Indian language revitalisation initiatives, educators and language workers reported persistent flaws: inaccurate translations, prioritisation of dominant languages, and erasure of culturally embedded meanings.

    “Educators and language workers reported persistent flaws: inaccurate translations, prioritisation of dominant languages, and erasure of culturally embedded meanings (Pradhan & Dey, 2025).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Pradhan & Dey, 2025

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Findingcitation

    In the Rooted in the Collective project, the AI systems discussed were hypothetical; generative AI appeared only as a speculative future for environmental monitoring and cultural storytelling.

    “While the design encouraged community ownership of data collection and critique of automation, the AI systems discussed were hypothetical; generative AI appeared only as a speculative future for environmental monitoring and cultural storytelling, given its capacity to synthesise diverse data sources (Moudgalya et al., 2025).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Moudgalya et al., 2025 source

  • May combine two claims.

    Findingcitation

    Urban farmer workshops cultivated AI literacy through speculative exercises, yet participants emphasised enduring risks of 'data apartheid'.

    “Urban farmer workshops cultivated AI literacy through speculative exercises, yet participants emphasised enduring risks of 'data apartheid' (Moudgalya et al., 2025).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Moudgalya et al., 2025 source

  • May combine two claims.

    Findingreasoning

    While co-creation under constraint generated valuable pedagogical adaptations, these efforts rarely altered systemic architectures.

    “Third, while co-creation under constraint generated valuable pedagogical adaptations, these efforts rarely altered systemic architectures.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingreasoning

    These cases reveal a sovereignty spectrum-from protocol-driven research governance in Abundant Intelligences to contingent co-creation in Brazil and India-indicating that where legal orders and relational protocols anchor participation, harms diminish and futures become thinkable.

    “These cases reveal a sovereignty spectrumfrom protocol-driven research governance in Abundant Intelligences to contingent co-creation in Brazil and Indiaindicating that where legal orders and relational protocols anchor participation, harms diminish and futures become thinkable.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The seven selected cases are representative of GenAI in Indigenous educational settings.
    • Rests on an assumption: The cases are comparable enough to support causal claims about sovereignty, legal orders, relational protocols, and harms.

  • May combine two claims.

    Findingcitation

    Indian language initiatives faced barriers: proprietary systems and translation inaccuracies closed off pathways for Indigenous governance, reinforcing linguistic hierarchies under corporate oversight.

    “Indian language initiatives faced similar barriers: proprietary systems and translation inaccuracies closed off pathways for Indigenous governance, reinforcing linguistic hierarchies under corporate oversight (Pradhan & Dey, 2025).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Pradhan & Dey, 2025

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Findingcitation

    Participants in the Rooted in the Collective project raised concerns over cost, access, and the risk of 'data apartheid' under corporate agricultural data regimes, illustrating how participatory learning can deepen critical awareness without enabling infrastructural sovereignty.

    “Participants raised concerns over cost, access, and the risk of 'data apartheid' under corporate agricultural data regimes, illustrating how participatory learning can deepen critical awareness without enabling infrastructural sovereignty (Moudgalya et al., 2025: 795).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Moudgalya et al., 2025 source

  • May combine two claims.

    Interpretationcitation

    Generative AI operates less as a deployed tool than as a shaping paradigm, influencing language technology futures in ways that entrench dependency unless sovereignty is embedded from inception.

    “Here, generative AI operates less as a deployed tool than as a shaping paradigm, influencing language technology futures in ways that entrench dependency unless sovereignty is embedded from inception (Pradhan & Dey, 2025).”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

    Interpretationreasoning

    True decolonial GenAI cannot end with bias correction alone; it must be structurally reimagined and grounded explicitly in Indigenous legal orders, ethical frameworks, and technological priorities.

    “Furthermore, true decolonial GenAI cannot end with bias correction alone.It must be structurally reimagined and grounded explicitly in Indigenous legal orders, ethical frameworks, and technological priorities.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    Indigenous governance of GenAI must extend beyond data inclusion to encompass full epistemic jurisdiction, including the right to withhold participation in extractive digital systems.

    “We argue that Indigenous governance of GenAI must extend beyond data inclusion to encompass full epistemic jurisdiction, including the right to withhold participation in extractive digital systems.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationcitation

    GenAI models trained predominantly on Western datasets inevitably reproduce these biases, positioning Western worldviews as objective and authoritative while rendering Indigenous contributions marginal or invisible.

    “GenAI models trained predominantly on Western datasets inevitably reproduce these biases, positioning Western worldviews as objective and authoritative while rendering Indigenous contributions marginal or invisible (Benjamin, 2019).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Benjamin, 2019

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    GenAI as Ceremony reframes AI not as a neutral tool nor merely a pedagogical aid, but as a participant in relational knowledge systems.

    “At its centre is GenAI as Ceremonya conceptual core that reframes AI not as a neutral tool nor merely a pedagogical aid, but as a participant in relational knowledge systems.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Interpretationcitation

    Indigenous Legal Orders are juridical infrastructures that enforce rights such as FPIC, ensuring these are ongoing, revocable, and codified in law rather than reduced to rhetorical gestures.

    “Indigenous Legal Orders: juridical infrastructures that enforce rights such as Free, Prior, and Informed Consent (FPIC), ensuring these are ongoing, revocable, and codified in law rather than reduced to rhetorical gestures (UN, 2007;Walter et al., 2020).”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites UN, 2007

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites Walter et al., 2020 source

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductnone given

    Iterations continued until no new eligible cases emerged, indicating conceptual saturation.

    “Iterations continued until no new eligible cases emerged, indicating conceptual saturation.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    DAIS moves beyond inclusion to insist on jurisdiction: Indigenous authority over infrastructure, datasets, and governance as a precondition for justice-extending from dataset licensing to mechanisms such as kill-switch mechanisms to disable models.

    “DAIS therefore moves beyond inclusion to insist on jurisdiction: Indigenous authority over infrastructure, datasets, and governance as a precondition for justiceextending from dataset licensing to mechanisms such as kill-switch mechanisms to disable models.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    Reverse Tutelage includes Indigenous-led parameter setting, evaluation metrics, and veto power.

    “This includes Indigenous-led parameter setting, evaluation metrics, and veto power.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    DAIS complements critical pedagogy by pairing critique with jurisdiction, ensuring pedagogical agency is backed by structural power.

    “DAIS complements critical pedagogy by pairing critique with jurisdiction, ensuring pedagogical agency is backed by structural power.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    Posthumanist perspectives decentre the human and emphasise entanglement between humans and technologies.

    “Posthumanist perspectives decentre the human and emphasise entanglement between humans and technologies (Barad, 2007;Snaza & Weaver, 2015).”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites Barad, 2007

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites Snaza & Weaver, 2015

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Classified as probably not a substantial claim. May combine two claims.

    Interpretationcitation

    Co-Creation in Education positions education as both site and method for rethinking AI futures-imagining oral interface models that respect ceremonial protocols or AI systems mediating land-based pedagogy under Indigenous governance.

    “Co-Creation in Education: positions education as both site and method for rethinking AI futuresimagining oral interface models that respect ceremonial protocols or AI systems mediating landbased pedagogy under Indigenous governance (Lewis et al., 2024;Simpson, 2014).”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites Lewis et al., 2024

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites Simpson, 2014 source

  • Classified as probably not a substantial claim.

    Study design and conductown data

    The DAIS framework emerged inductively from the comparative analysis of these cases rather than being imposed a priori.

    “Building on these insights, the DAIS framework emerged inductively from the comparative analysis of these cases rather than being imposed a priori.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The comparative analysis of secondary sources is sufficient to develop a normative framework for Indigenous AI sovereignty.

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    Addressing bias in GenAI requires more than inclusion or dataset diversification-it demands Indigenous jurisdiction over the epistemic, legal, and material conditions under which AI systems operate.

    “Addressing it requires more than inclusion or dataset diversificationit demands Indigenous jurisdiction over the epistemic, legal, and material conditions under which AI systems operate.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

    Interpretationreasoning

    DAIS shifts the field from bias correction towards jurisdiction, from optimisation towards ceremony, and from inclusion towards Indigenous self-determination.

    “In so doing, it shifts the field from bias correction towards jurisdiction, from optimisation towards ceremony, and from inclusion towards Indigenous self-determination.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    Open-source availability is insufficient without Indigenous-controlled stacks for data storage, access, and model deployment.

    “For infrastructure, open-source availability is insufficient without Indigenous-controlled stacks for data storage, access, and model deployment.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    DAIS affirms relationality but anchors it in enforceable governance-consent as ongoing practice, legal authority over data, and the right of refusal.

    “DAIS affirms relationality but anchors it in enforceable governanceconsent as ongoing practice, legal authority over data, and the right of refusal.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    DAIS reframes AI's role in education: not as a neutral tool or mere collaborator, but as a ceremonial participant governed by Indigenous law.

    “DAIS reframes AI's role in education: not as a neutral tool or mere collaborator, but as a ceremonial participant governed by Indigenous law.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Study design and conductcitation

    The design enables interpretive comparison across diverse cases to identify convergent and divergent patterns while preserving contextual nuance.

    “This design enables interpretive comparison across diverse cases to identify convergent and divergent patterns while preserving contextual nuance in theorising governance challenges and possibilities for Indigenous AI futures.”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites Cruzes et al., 2015

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites Yin, 2018

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductcitation

    Themes were generated inductively and refined through investigator triangulation, enhancing rigour and reducing interpretive bias.

    “Themes were generated inductively and refined through investigator triangulation, enhancing rigour and reducing interpretive bias (Braun & Clarke, 2016).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Braun & Clarke, 2016

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    By centreing enforceable sovereignty, DAIS reframes what ethical AI governance requires.

    “By centreing enforceable sovereignty, DAIS reframes what ethical AI governance requires.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    Fairness becomes derivative, not primary.

    “Fairness becomes derivative, not primary.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    Where posthumanism is philosophical, DAIS is juridical and infrastructural.

    “In short, where posthumanism is philosophical, DAIS is juridical and infrastructural.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationcitation

    AI systems marginalise non-Western ways of knowing, reinforcing settler futurity.

    “Figure: highlight how AI systems marginalise non-Western ways of knowing, reinforcing what Tuck and Yang describe as settler futurity; the projection of colonial worldviews into technological futures.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Tuck and Yang

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Interpretationreasoning

    DAIS stands as a counterpoint to UNESCO (2021) and OECD (2019) AI ethics frameworks, which prioritise inclusion, transparency, and accountability but leave colonial infrastructures intact by omitting jurisdictional authority.

    “At the global level, DAIS also stands as a counterpoint to UNESCO (2021) and OECD (2019) AI ethics frameworks, which prioritise inclusion, transparency, and accountability but leave colonial infrastructures intact by omitting jurisdictional authority.”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites UNESCO (2021)

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites OECD (2019)

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationcitation

    Epistemic Sovereignty affirms Indigenous jurisdiction over how knowledge is created, shared, and interpreted.

    “Epistemic Sovereignty (cross-cutting foundation): affirms Indigenous jurisdiction over how knowledge is created, shared, and interpreted.”
    Quote in the paperNot a substantial claim
48 of 48

  • The paper relies on this without support.

    Contemporary GenAI systems increasingly draw on Indigenous languages, ecological knowledge, and oral traditions to refine their outputs.

    “Like earlier extractive economies that dispossessed Indigenous peoples of land, contemporary GenAI systems increasingly draw on Indigenous languages, ecological knowledge, and oral traditions to refine their outputs.”

  • The paper relies on this without support.

    The seven selected cases are sufficient to demonstrate the DAIS model and cross-case patterns.

    “Seven cases, summarised in Table 1, met the criteria and constitute the analytic corpus.”

  • The paper relies on this without support.

    The seven selected cases are representative of GenAI in Indigenous educational settings.

    “Across the seven cases, GenAI in Indigenous educational settings appears as both possibility and risk.”

  • The paper relies on this without support.

    The comparative analysis of secondary sources is sufficient to develop a normative framework for Indigenous AI sovereignty.

    “Building on these insights, the DAIS framework emerged inductively from the comparative analysis of these cases rather than being imposed a priori.”

  • The paper relies on this without support.

    The cases are comparable enough to support causal claims about sovereignty, legal orders, relational protocols, and harms.

    “These cases reveal a sovereignty spectrumfrom protocol-driven research governance in Abundant Intelligences to contingent co-creation in Brazil and Indiaindicating that where legal orders and relational protocols anchor participation, harms diminish and futures become thinkable.”
5 of 5

Cites

Flagged first, then by citations

Research works printed but not linked: 5

  1. Local Contexts. (2019). Traditional knowledge (TK) labels. https://localcontexts.org/labels/traditional-knowledge-labels/
  2. Miles, M. B., Huberman, A. M., & Saldaña, J. (2014). Qualitativedata analysis: A methods sourcebook. (3rd ed.) SAGE.
  3. 10.4324/9780429273957 · DOI · Paper page
  4. Wilson, S. (2008). Research is ceremony: Indigenous research methods. Fernwood Publishing. https://fernwoodpublishing.ca/book/research-is-ceremony
  5. Yin, R. K. (2018). Case study research and applications: Design and methods. (6th ed.). SAGE.

Cited by

Sorted by citations

Related

Authors
Mbah, M. F., & Jarvis, L.
Affiliations
University of Manchester (GB)
Corresponding author
Mbah, Marcellus Forh
ORCID
Mbah, Marcellus Forh
Venue
Cogent Education
Publisher
Informa UK Limited
Published
2026-02-12
Volume
13
Issue
1
ISSN
2331-186X (ISSN-L)
DOI
10.1080/2331186x.2026.2626631
OpenAlex
W7128728847
Cited by
7
References
43
Citations by year
2026: 7
FWCI
51.06
Type
journal article
Language
en
Open access
Licence
Locations
  • Cogent Education (publishedVersion, other-oa)
  • Research Explorer (The University of Manchester) (publishedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
University of Manchester (Humanity Strategic Fund)
Topics
  • Social Sciences › Health › Indigenous Health, Education, and Rights
  • Social Sciences › Safety Research › Ethics and Social Impacts of AI
  • Physical Sciences › Computer Science › Information Systems › Digital Education and Society
Keywords
generative AI, indigenous education, Indigenous data sovereignty, epistemicide, colonial hierarchies, relational ethics, AI ethics, Indigenous resurgence

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-30.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-30.

Cited sources are checked by gpt-6-sol.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review