Listening to the conversations around the United Nations General Assembly and Climate Week, I kept returning to a question that follows me into every AI workshop I facilitate: who gets to decide?
Who decides which problems deserve investment? Whose knowledge counts? Who carries the risk when a promising tool fails? And what power do the people affected have to change what happens next?
In an interview with Vanguard Charitable, Project Evident’s Kelly Fitzsimmons describes AI as “an arrival technology, like electricity.” She calls on donors to provide the unrestricted support nonprofits need to experiment, develop skills, and participate.
Her analogy stayed with me. Electricity’s arrival did not bring everyone onto the grid. Extending it across rural America required public financing, infrastructure, and institutions, including cooperatives through which communities organized to secure service.
Participation had to be built.
In 2024, I coauthored a piece for CEP calling on philanthropy to fund the infrastructure, shared guidelines, and equitable innovation that responsible AI requires. Fitzsimmons’s call for unrestricted support reinforces that argument: nonprofits need resources to learn, experiment, and pursue better outcomes. I want to carry the conversation one step further. Meaningful participation also means helping define those outcomes, setting boundaries, challenging decisions, and having the authority to say that a particular use should not proceed.
That is a useful place to begin this conversation.
Access to a tool does not automatically give us the capacity to use it well, the authority to shape its use, or the power to challenge what happens because of it.
Philanthropy can help build those conditions. We have work to do within our organizations, where AI is already entering everyday decisions. We also have a responsibility to support public accountability for the systems we do not control.
Both require resources. Both require us to listen. And both require something to change because we did.
What support makes possible
The new survey from NTEN and The Bridgespan Group gives us a picture of work already underway and support still catching up.
Among respondents answering the question about everyday use, roughly three-quarters use AI at least weekly. Among executive respondents, only 17 percent report a designated AI budget in place, 22 percent report staff training on safe and responsible use in place, and 22 percent report an AI risk-management plan in place.
Many organizations are trying to build these supports. Roughly 44% report responsible-use training in development, and 45% have risk-management plans in development. There is effort and plenty of unfinished work.
Read those findings next to CEP’s State of Nonprofits 2026. Forty-six percent of nonprofit CEOs say their own burnout is very much a concern. Seventy-three percent report increased demand for services, and 57% say foundation grants have become harder to secure since January 2025.
We are asking people under extraordinary pressure to make consequential technology decisions while sustaining the work their communities already depend on.
A subscription and an invitation to innovate leave much of that responsibility under-funded.
But offering a program does not necessarily solve the problem either. In her recent CEP article, NewSchools Venture Fund CEO Frances Messano points to another finding: 46% of nonprofit CEOs say they do not have the bandwidth to absorb new capacity-building support.
That should give us pause.
A cohort, training series, or planning process may be useful. It may also be another thing to fit into an already impossible week. Before offering AI support, we need to ask what would make responsible use possible for this organization.
The answer might be coaching, training, or advisory support. It might be flexible dollars, funded staff time, or fewer reporting requirements.
Let what we hear impact what we offer.
And make it safe to tell us the truth. Messano cites research finding that only 41% of nonprofit CEOs feel they can be completely honest with their funders. If we are going to successfully drive change, we need our nonprofit partners need to be able to say that a tool failed, that they lack capacity, or that they do not want to use AI without fearing their honesty will count against them.
Where the judgment shifts
Consider a foundation using AI to synthesize community feedback.
This is a possibility I find exciting. Much of my work on oral and alternate reporting has been about taking qualitative knowledge seriously: conversations, observations, stories, and existing materials that do not fit neatly into a funder’s reporting template.
AI may help us learn across more of that information. It may also flatten disagreement, erase context, or give a funder’s preferred interpretation the appearance of objectivity.
Who checks whether the summary represents what people meant? Who decides which disagreements matter? Can the people whose experiences informed the analysis challenge its conclusions before those conclusions influence funding?
This is where governance becomes tangible.
The object of governance is the work in which AI participates.
When I ask, “Where did the judgment shift?” I am asking us to notice when a summary becomes an assessment, an assessment becomes a recommendation, or a recommendation begins to determine someone’s access to resources.
As consequence increases, governance increases.
We need to name what AI is doing, what remains a human responsibility, who has authority to act, and what would require us to stop. Human oversight requires someone with the knowledge, time, and power to question an output and change the outcome.
Those conditions belong in the budget and the design of the work.
Start with a workflow people recognize. Make the decisions visible. Give them somewhere to bring uncertainty. Test, learn, and revise.
Participation also needs to extend beyond contributing information. People deserve to know what we heard, what changed, and what we decided against. Their words becoming part of an analysis is not the same as their having influence over a decision.
Knowledge should return value to the people who shared it.
Who governs the builders?
There are limits to what we can accomplish inside our own organizations.
A nonprofit can establish careful rules for handling information, checking outputs, and authorizing actions. It cannot resolve failures of control at an AI company through its responsible-use policy. It cannot make a provider’s safety claims trustworthy simply by training its staff.
Providers must remain accountable for the systems they build. Independent scrutiny and public oversight need resources and authority.
The social sector has knowledge that belongs in this work. Organizations working on education, health, housing, civil rights, and the environment understand consequences that may be overlooked. That knowledge needs a way to influence development, evaluation, and public decisions.
Philanthropy can fund independent research, public-interest advocacy, and institutions that help communities scrutinize claims and challenge harmful practices. It can pay people for their participation and invest in the expertise they need to exercise meaningful influence.
A nonprofit’s responsibility to use AI carefully and a funder’s responsibility to build accountable systems must develop together. That requires resources.
Fund the work
Within an organization, begin with one consequential workflow.
Ask the people doing and experiencing the work where AI is already participating, what would help, and what needs to remain human. Establish who can decide and who is accountable. Pay for the staff time, learning, and evaluation needed to test the approach and decide whether to proceed, change course, or stop. Make honesty safe by allowing partners to tell us when an experiment fails or a requirement creates more work than value. Then change our decisions in response.
Beyond our organizations, fund the independent researchers, journalists, advocates, and community institutions that can investigate providers and challenge harmful practices. Pay community members for their time. Support them in building expertise, setting priorities, and shaping decisions.
Judge that investment by what changes: whether staff have room to learn and exercise judgment, whether communities can challenge decisions made in their name, and whether their participation changes what happens next.
I remain interested in what AI might help us repair. But hope needs a budget. Staff time, independent scrutiny, and community participation all cost.
The technology is here. The terms of participation are still being written. Philanthropy has the resources to help more people shape them.
Fund the people. Fund the learning. Fund the institutions that hold power accountable. And give the people closest to the consequences enough authority to change what happens next.
Sources and further reading
Vanguard Charitable. “Building for the future: The donor’s role in nonprofit modernization.” Interview with Kelly Fitzsimmons, founder and CEO of Project Evident, June 2026.
USDA Rural Development. “USDA Celebrates 90th Anniversary of Rural Electrification Act, Billions Invested in Electric Infrastructure.” Background on public financing and rural electric cooperatives.
NTEN and The Bridgespan Group. State of Nonprofit AI Adoption and Governance. September 2026. The everyday-use figures and organizational capacity figures reflect different respondent groups.
Center for Effective Philanthropy. State of Nonprofits 2026: What Funders Need to Know. May 2026.
Frances Messano. “Listening Isn’t Enough, Philanthropy Has to Respond.” Center for Effective Philanthropy, October 1, 2026. Discusses separate CEP research on nonprofit needs and foundation responses.
Rachel Kimber, Joanna Drew, Ravit Dotan, and Mark Greer II. “Responsible AI: How Philanthropy Can (and Should) Support the Movement.” Center for Effective Philanthropy, March 5, 2024.
Research note: NTEN’s weekly-use figure reflects 723 staff and executive respondents; the organizational capacity and governance figures reflect approximately 404 executive respondents. These are different respondent groups. CEP’s State of Nonprofits 2026 describes U.S. nonprofits receiving foundation funding. The findings cited through Frances Messano’s article come from separate CEP research on foundation responses and nonprofit needs.