In much of impact investing, AI is being treated as no longer optional. New tools promising faster diligence, cleaner reporting, and more efficient capital deployment arrive each week. Many investors have moved past asking whether to use AI and are now asking which tool first.
At Raven Indigenous Outcomes Fund, that question is still live. We do not assume the answer is always yes.
We are an Indigenous-led impact investor working alongside communities across what is now North America. Our work is grounded in self-determination, community wellbeing, and responsibilities to land, to people, and to the generations ahead. We and other impact investors are working toward a different economic system, one that puts community, land, and long-term wellbeing ahead of individual wealth.
That work is real, but it is hard and slow. AI is not waiting and we do not have the luxury of staying outside the present system while we work toward a better one. What we can do is choose, openly and deliberately, where we engage, where we refuse, and what we say plainly about both.
Our position: AI should be tested case by case, against the same net impact standard we apply to every other investment and instrument we use, with our marginal contribution to the harms weighed against our marginal contribution to the benefits. Not on momentum. Not on principle. On honest accounting, without losing sight of our long-term objectives
The harms are real
Any honest conversation about AI in Indigenous-led impact begins with its footprint on Mother Earth, on our communities, and on our relations.
AI is not weightless. It is built on data centres, power grids, water systems, and mineral supply chains. The International Energy Agency projected, in 2025, that global data center electricity demand would more than double by 2030, with AI driving most of that growth. We are now in that future. Those systems draw heavily on territories already under pressure, including lands stewarded by Indigenous nations.
This pattern is familiar. Resources taken from one place to create value in another. Costs that fall on Mother Earth, on community, and on future generations, never evenly shared. From our perspective, this is not only an environmental concern. It is a question of consent and relationship. Who was asked? Who benefits? Who carries the burden? Who is left to live with the consequences?
The harms extend into knowledge. AI trained on dominant data sources reproduces dominant worldviews. Indigenous knowledge, when it appears at all, is taken out of context, generalized across nations, or treated as fixed rather than living and place-based. AI systems have already been documented generating fabricated cultural information, undermining knowledge meant to be held, practised, and passed on with care.
A further harm remains largely invisible: the global labour required to train and maintain these systems, often under conditions no impact portfolio would tolerate elsewhere.
These are not edge cases. They are part of how the economic system works today.
Why disengagement fails
We are part of an economic system that focuses on extraction and values speed, scale, and cost reduction. That system did not ask us before adopting AI, and it will not pause because we have concerns. AI is already shaping funding processes, regulatory expectations, and the ways community-led projects are assessed. Communities will encounter these expectations whether or not we engage with AI ourselves.
Stepping away from AI does not free us from it. It only removes a tool we could choose to use, on community terms.
Our approach to investing has never been about avoiding all negative impacts. That is unachievable. It has been about asking whether a project has a net positive contribution, over time, to the wellbeing of our communities and the restoration of a reciprocal relationship with Mother Earth, and with one another. We test each investment against that belief rather than assume compliance. We walk away when the test fails.
Like any tool, AI requires rigorous scrutiny. Because it evolves so quickly, we must constantly question why we use it, how we use it, and at whose expense.
Net impact in practice
A net impact lens applied to AI is not an assessment of AI as a global system. We do not claim credit or blame for the full footprint of an industry we did not build. From our perspective, the relevant calculation is our marginal contribution.
When we use AI in a specific way, we ask:
- What does our use contribute to the harms? Incremental electricity and water demand at data centres. Downstream signal supporting further AI development. Displacement of work that would otherwise be paid. Exposure of community knowledge to systems we do not fully control.
- What does our use contribute to the benefits? Faster, more accessible processes for community planning. Stronger documentation and negotiation capacity on the community side. Lower cost of capital access at community scale. Tools and templates that build capacity beyond a single project.
- Do the benefits we contribute outweigh the harms we contribute, in this specific case, for these specific communities?
The test does not always produce comfortable answers. It is meant to produce honest ones.
One use case
One place the test could hold is in helping communities articulate, structure, and document their own impact thinking.
This work sits inside a broader push for us, well underway before AI entered the conversation, to build capacity with the communities we work alongside by sharing tools, templates, frameworks, and process support.
The premise is straightforward: capital access is often gated less by the quality of community projects than by the documentation conventions of capital providers, including most impact investors. Closing that gap, without changing the substance of what communities want to do, is what most of our capacity building is for.
We are building an impact framework tool that fits inside this work. Conceptually, it helps a community move from vision to a structured set of measurable outcomes, holding financial returns and community-defined indicators of wellbeing in view together. It surfaces relevant evidence. It suggests measurement and valuation approaches. It structures output in formats capital providers expect.
The tool lays out knowledge, best practices, and structures that fit the economic system’s expectations, so that communities can engage that system from a position of clarity and access rather than from a position of translation cost. It does not decide. A human on the community side leads every decision.
The reason we believe the test holds here:
- Our marginal contribution to harm. The full harm list (environmental footprint, data sovereignty risk, and displacement of paid labor) applies here, but each is bounded. The environmental footprint is bounded by the realistic alternative: consultant-driven work has its own substantial footprint, so our marginal addition is small. Data sovereignty risk is bounded by keeping community data within agreed scope. Labor displacement is bounded by the tool not replacing paid work; we continue to fund the human capacity around it on community terms.
- On the benefit side. Documentation that took months becomes iterative work done in weeks. The cost of producing the conventions that gate capital comes down. Communities engage in capital conversations using the language of the institutions across the table. More capital flows to community-led projects, funding economic and social development on community terms, and deepening community wellbeing for the generations ahead.
- What still gives us pause. Whether the tools shape community thinking in ways we did not intend. Whether the time saved goes to the community or back to capital providers. Whether what holds early holds at scale. Most importantly, we must question whether we are contributing to further assimilation into the dominant system, rather than changing the system for a more inclusive, values-aligned one.
Other applications may pass the same test and will be examined in the same way. In each case, the outcome depends on how the tool is used, how it’s governed, and whether its use strengthens or weakens community self-determination.
Ownership, Control, Access, and Possession
The OCAP principles (Ownership, Control, Access, and Possession) were developed by the First Nations Information Governance Centre and remain a critical guide for how data should be governed. They were not designed for AI, but they reflect something deeper: the right of communities to steward their own knowledge and information.
In practice, perfect alignment is rare. Many tools already in use across the impact sector, from cloud document systems to common analytics platforms, involve partial trade-offs in community ownership, control, and access. AI introduces new versions of those trade-offs, sometimes more pronounced, but the underlying tension is not new.
We owe it to the community to be honest about our limitations before we begin working together. We use OCAP as a set of questions:
- Who owns this data?
- Who decides how it is used?
- Where does it live?
- What is lost, and what is gained, by using this tool?
Informed consent is the floor. The standard we want to hold ourselves to is ongoing, mutual, and revisable. Communities should be able not only to understand the implications of a tool at the start, but to change terms as work evolves, and to withdraw if something is not working. That is closer to what right relationship asks of us: not a one-time disclosure, but a continuing accountability.
Our commitments
We are not offering a final answer. We are committing to a way of working.
- Apply the marginal net impact test, in writing, before any new AI use begins.
- Make trade-offs visible to communities and to the field, including where we remain uncomfortable with our own decisions.
- Treat consent not as a single checkbox, but as an active, revisable relationship that honors community ownership and self-determination.
- Keep asking, as the technology evolves, why we are using it and how. Walk away when balance fails.
For the impact investing field, this invites a shift in the question being asked. Stop asking which AI tool to adopt next. Start asking whether AI use, in any specific case, is grounded in responsibility towards people and the planet, in consent, and in care.
For Indigenous-led organizations, engagement with AI is not endorsement. We can use these tools while flagging their costs and refuse them when our use would deepen the harms. What is required is clarity about where lines are drawn, and the courage to draw them in public.
Technology exists to serve relationships, between people, and between people and Mother Earth. Communities decide which relationships are worth sustaining. Technology follows.
