Key Takeaways
- Research argues that Indigenous communities should control how AI systems collect, interpret, and apply their knowledge
- CARE, OCAP®, and UNDRIP offer governance foundations for consent, accountability, and collective benefit
- Collaborative workshops, community-curated repositories, and language projects show alternatives to extractive AI development
Generative and agentic artificial intelligence are raising familiar concerns about job displacement and automated decision-making. For Indigenous communities, however, the stakes extend beyond whether machines replace particular tasks. AI systems can absorb languages, cultural records, ecological observations, biometric information, and geospatial data, then reuse that material far beyond its original context.
A growing body of research establishes that the issue is not solved by putting more Indigenous knowledge into conventional models. The consequential question is who controls collection, interpretation, access, and deployment. Embedding Indigenous Data Sovereignty in Ethical AI Governance, published in 2026, identifies Indigenous data sovereignty and the CARE Principles as foundations for culturally aligned AI governance.
CARE stands for Collective Benefit, Authority to Control, Responsibility, and Ethics. Its emphasis differs from governance programs centered mainly on individual privacy or regulatory compliance. Indigenous knowledge is often collective, relational, and connected to specific territories, communities, or responsibilities. Permission from one person may not establish legitimate authority to train a model on material held by a people.
Data can be technically public while remaining culturally restricted. A language recording, map, medicinal account, or ceremonial description may be available online without being appropriate for model training. Standard web-scraping practices rarely capture those distinctions.
That governance gap has prompted researchers to use the term "AI extractivism." Preventing AI extractivism: the case for braiding indigenous knowledge into AI examines the risks of harvesting linguistic, biometric, geospatial, and ecological information without adequate consent, compensation, or accountability. The pattern resembles earlier forms of extraction: value moves outward, while control and long-term benefits remain elsewhere.
For enterprises, Indigenous data sovereignty operates as a strict product governance and procurement requirement. Data lineage records must document collective authority, usage conditions, cultural restrictions, benefit-sharing arrangements, and withdrawal procedures. Model cards and privacy notices alone rarely capture those obligations. Agentic systems add another layer because they can retrieve, combine, summarize, and act on information across connected repositories.
Participatory work described in Towards Participatory AI with ovaHimba and San People in Namibia points toward community-curated digital repositories and culturally valid design processes as alternatives to extraction. Rather than transferring knowledge into an externally governed database, communities shape repository structure, decide what belongs there, and define who may use it. This changes both the system architecture and the consultation process.
TELUS's Indigenous AI workshops offer a practical example of this approach. A 2025 humanitarian-technology study involving TELUS and Indigenous communities emphasized participation across the full AI lifecycle, from early design choices through ongoing governance. A consultation near launch is not equivalent to shared decision-making over objectives, training data, evaluation methods, access, and oversight.
Projects in Northeast India also illustrate how AI assists with language and ecological knowledge preservation. Yet preservation is not automatically protection. Digitization makes vulnerable knowledge easier to copy, translate, commercialize, or detach from its cultural setting. Community-defined permissions and technical access controls therefore belong in the core architectural planning.
OCAP®, meaning Ownership, Control, Access, and Possession, provides an additional governance framework, particularly in First Nations contexts. Alongside CARE, it complements the United Nations Declaration on the Rights of Indigenous Peoples, enforcing free, prior, and informed consent and self-determination. These approaches frame governance as an ongoing relationship rather than a single compliance checkbox.
Implementation requirements vary among Indigenous peoples. Enterprises must avoid turning one framework into a universal template or assuming that a single representative can authorize every use, as local governance structures, laws, protocols, and expectations differ significantly.
Organizations building AI for environmental monitoring, healthcare, telecommunications, education, mapping, language services, or public administration often encounter Indigenous knowledge even when it is not explicitly labeled. Early community participation, traceable permissions, purpose limits, compensation, and enforceable access rules reduce legal and reputational exposure. Furthermore, these practices shift AI development from knowledge extraction toward partnerships where Indigenous communities retain meaningful authority over their own data and science.
⬇️