Indigenous and community data sovereignty

A community — Indigenous nations, kin networks, diaspora associations, language communities — asserts ownership and governance over data generated by or about its members: language corpora, genomic samples, traditional-knowledge records, demographic statistics, place-name datasets. The work institutionalises that ownership through community data trusts, sovereignty principles (the CARE principles for Indigenous data governance), community-controlled ethics review, and licensing regimes that require ongoing community consent and benefit-sharing for any AI use of the data. Where a state or company has already taken the data, the work becomes restitution: deletion, repatriation, retraining without it.

An actor chooses this strategy because AI's central feedstock is data, and the default extractive pipeline routes that data through state and corporate intermediaries who treat communal knowledge as a public good available for free. A sovereignty framing reasserts that data carries ongoing rights of its source community — not merely the rights of individuals inside it — and shifts the unit of consent from the individual to the collective. The approach also provides a mechanism communities can actually run themselves, which is rare in a movement where most leverage requires access to legislatures, courts, or capital the community does not control.

It trades interoperability with the dominant infrastructure for governance autonomy. A sovereignty regime forces a fork in the technical stack — separate AI tools trained on community-governed corpora, separate evaluation procedures, separate hosting — which is expensive to build and difficult to sustain at the scale a community of a few hundred thousand people can fund. The strategy also depends on state recognition of communal data rights, which most jurisdictions grant unevenly, and is fragile against companies that argue their AI was already trained on the extracted data before the sovereignty regime existed.

Verdict — strong at the local-institutional and international-norms scales, structurally limited at the frontier-lab and beyond-CANZUS scales

The strategy earns its keep where the work product is either a community's own AI infrastructure built under its own licence or a shared international-norms register that adjacent institutions adopt as their data-governance frame. At the first scale, the Te Hiku Media Kaitiakitanga License campaign is the corpus's clearest case: a Māori iwi-broadcaster building first-rate te reo Māori automatic-speech-recognition (92 percent accuracy) under a community-set stewardship licence, with multi-year flagship state funding (NZ$13M MBIE Strategic Science Investment Fund placement) flowing without requiring relaxation of the sovereignty terms. The strategy here works as a capability-enabling apparatus, not a capability-compromising one — the Kōrero Māori crowdsource refuted the standing objection that restrictive licensing terms deter the volunteer contribution needed for credible corpus-building. At the second scale, the Global Indigenous Data Alliance and the CARE Principles operate as the international register: a peer-reviewed scholarly codification, a "#BeFAIRandCARE" institutional shorthand that adjacent data-governance infrastructure can adopt without bilateral negotiation, and a 2026 Texcoco AI workshop that converted the framework from a research-data-management norm into a working register for Indigenous-AI protocols. The framework's third operative form — the First Nations Information Governance Centre's OCAP® trademark — converts a normative principle into a legal claim FNIGC can enforce against research-body misappropriation in Canadian law.

The strategy weakens sharply at two further scales the corpus's signal makes legible. First, at the commercial frontier-AI lab scale: no documented case of a frontier lab agreeing to retrain a deployed model after community objection appears in the corpus's signal. The strategy wins concessions at the research-consortium and dataset-curator level (academic publishers adopting CARE, IRBs requiring OCAP®-aligned protocols) but has not pierced the frontier-lab tier where AI training is concentrated. Second, at the beyond-CANZUS scale: GIDA's institutional spine is heavily settler-colonial-state-located (Australia, Aotearoa NZ, US, with Canada via FNIGC, with GIDA-Inuit and GIDA-Sápmi added more recently); coverage in Latin America, sub-Saharan Africa, the Pacific island nations beyond Aotearoa, and Southeast Asia is thinner. Indigenous-language data outside the CANZUS frame is subject to the same web-scraping under the same conditions, but the operative enforcement infrastructure is much shallower. The strategy's most concrete unfinished business is making the sovereignty terms binding inside the dominant infrastructure (frontier labs that have already trained on the data) and extending the institutional infrastructure beyond the CANZUS spine (the Aotearoa Treaty-of-Waitangi context where the strategy currently runs strongest is not, by itself, transferable to jurisdictions without comparable Crown-Indigenous reciprocity).

The strategy's most distinctive single finding from this depth pass is that its case is strongest exactly where it is operationally narrowest: where a small community holds its own data, builds its own AI, and runs its own licence, with state cooperation that is the historical exception rather than the rule. Whether the framework can scale — beyond the rohe of speakers small enough that a multi-decade archive suffices, beyond settler-colonial jurisdictions willing to recognise treaty-based communal claims — is the strategy's open empirical question, and the corpus's signal does not yet provide a confirmed positive case.

Ecology

The strategy strongly overlaps with parallel community research institution: both build separate, community-governed infrastructure rather than seeking concessions from the dominant infrastructure. Te Hiku Media is itself a parallel-community-research-institution adopter operating under a data-sovereignty licence; FNIGC operates as a parallel community-research surveying body filling a void federal population surveys left open by exclusion. The two strategies are partly the same institutional form wearing different analytical hats: parallel-community-research-institution names the institutional form (an independent research body outside the dominant infrastructure), data-sovereignty names the governance principle under which that body operates. The cleanest distinction is that parallel-community-research-institution may run on conventional open-research norms (DAIR publishes openly under standard scholarly licensing) while data-sovereignty asserts a specific community-rights claim over upstream data that conventional open-science norms do not honour.

The strategy runs through counter-narrative framing as one of its load-bearing tracks. The framing itself — "Data is the last frontier of colonization", "I don't just think about how AI can be used, I think about how we can be the makers of AI", "be FAIR and CARE", "data as taonga" — is a sequence of reframings contesting the open-data movement's working assumption that data is a public good available for free. The Kaitiakitanga License contains a strategic-narrative component (the preamble's framing of digital colonisation, the essay's "wholesale appropriation of existing culture" claim) as much as a legal-instrument component; the message-side codification of the framing is the Indigenous data sovereignty message entry. The two strategies feed each other: each new institutional adoption of CARE or OCAP® reinforces the counter-narrative's legibility, and each propagation of the counter-narrative widens the institutional-adoption surface.

The strategy is adjacent to strategic litigation against algorithmic state decisions through their shared rights-claim register. Both ground claims in formal legal instruments — UNDRIP, treaty rights, the International Covenant on Civil and Political Rights for litigation; UNDRIP, treaty rights, the OCAP® trademark, and the CARE Principles for sovereignty. Both deploy expert evidence and constituency-witness testimony as their evidentiary forms. The two strategies share lawyers, witnesses, and rhetorical infrastructure but currently operate in separate institutional tracks: future litigation enforcing CARE Principles against an AI training dataset — a Foxglove- or AI-Now-style case applying the IDS framework as the legal cause of action — would be the cleanest fusion point and would convert the strategy from a governance-from-design move into a remedies-against-extant-deployments move. The fusion has not happened in the corpus's signal as of June 2026.

The strategy is adjacent to empirical audit and expose but operates in a distinct register. Audit-and-expose names harm in existing systems (Buolamwini's Gender Shades-anchored algorithmic-discrimination findings, Gebru's Stochastic Parrots critique of large-language-model training); data sovereignty asserts community authority over the upstream data, refusing the audit-then-fix model in favour of governance-from-design. The two strategies feed counter-narrative-framing through different lineages — audit-and-expose through algorithmic-accountability empirical findings, data sovereignty through Indigenous-rights normative claims — and converge in the broader decolonial-AI register the corpus tracks through the adjacent data colonialism and algorithmic colonialism framings.

The strategy contains but exceeds participatory deliberation as policy input, which Te Hiku Media and the Kaitiakitanga License campaign already carry as a documented adopter. Deliberation alone produces input to others' decisions — community submissions to state regulatory processes, hui contributions to research-ethics review, public consultation on policy drafts. Sovereignty transfers the decision authority itself — the Kaitiakitanga License is not a participatory-deliberation input to a research body's licensing decision; it is the operative licensing decision that the research body must accept or refuse. The strategy's depth verdict on participatory deliberation is that it is necessary but not sufficient: every sovereignty regime contains deliberation as a procedural component, but a deliberation regime without authority-transfer collapses back into consultative input the dominant infrastructure can decline.

The strategy's principal in-corpus adopters are the four organisational anchors — Te Hiku Media, Te Mana Raraunga, FNIGC, and GIDA — and the Kaitiakitanga License campaign. The corpus's coverage of the strategy is institutionally strong at the CANZUS spine, thinner outside it. The gap-loop work the strategy's analysis surfaces (a non-Māori community-stewardship licence applying the same logic to another indigenous language; a Latin American or sub-Saharan African or Pacific-island-beyond-Aotearoa adopter of the GIDA framework with its own institutional infrastructure; a documented frontier-lab concession achieved through sovereignty-framed pressure) is the principal motion the corpus's coverage of the strategy needs to deepen the institutional reach beyond its current settler-colonial-state concentration.