How NGOs, nonprofits, and schools are building something AI can’t fake: trust.
There’s a version of the AI story that goes like this: powerful technology arrives, institutions adopt it enthusiastically, things go quietly wrong, and nobody says anything until they have to. That story is playing out in a lot of places right now. But there’s another version — less covered, frankly more interesting — where organizations choose to be upfront about the technology they’re using, acknowledge its limitations honestly, and end up with stronger relationships with the people they serve because of it.
This piece is about that second story. About the NGOs, nonprofits, and educational institutions that are treating transparent communication not as a risk-management exercise, but as a genuine strategy for building something durable.
What “Bad AI” Actually Looks Like on the Ground
Before the wins, the context. AI in mission-driven sectors isn’t abstract. The risks are specific, and in some cases, already causing harm.
In higher education, predictive algorithms widely used by universities have been found to underestimate the success potential of Black and Hispanic students while overestimating that of white and Asian students, according to a 2024 study published in AERA Open. These tools are used to determine who gets academic support and resources. When the algorithm is wrong in a racially patterned way, it doesn’t just make an administrative error — it narrows the futures of real students.
The AI detection side has its own problems. A 2023 Stanford study found that seven leading AI detectors misclassified 61% of essays written by non-native English speakers as AI-generated, while almost no native-English essays were falsely flagged. In response, Yale, Johns Hopkins, Vanderbilt, the University of Waterloo, and at least a dozen other large universities have disabled or blocked Turnitin’s AI detection option entirely, citing bias and inaccurate accusations. That’s a significant institutional acknowledgment: we tried a tool, it hurt students unfairly, we stopped using it. That kind of honesty, stopping, naming the problem, changing course, matters.
In the nonprofit and humanitarian space, the risks are different but equally grounded. AI introduces risks including data privacy breaches, hallucinations, bias, and harm to trust with clients, donors, volunteers, and communities. For organizations that work with vulnerable populations, including but not limited to, refugees, domestic abuse survivors, people in food crisis — an AI system that misidentifies a need, or processes sensitive data without proper consent, isn’t a technology failure. It’s a safeguarding failure.
The Nonprofit Alliance is blunt about the regulatory backdrop in the US: the Trump administration’s 2025 executive order revoked Biden-era AI safety requirements and created bureaucratic roadblocks for AI regulation, while leaving organizations largely on their own to figure out ethical standards. That’s not a comfortable position for organizations whose entire value proposition rests on being trustworthy.
The Organizations Deciding to Say the Quiet Part Out Loud
NetHope and the Humanitarian Sector’s Public Ethics Infrastructure
NetHope — a consortium of around 60 global nonprofits — has probably done more than any single organization to create shared AI ethics infrastructure in the humanitarian sector. They launched an AI Working Group in 2019, long before the current wave of enthusiasm. Their AI Ethics for Nonprofits toolkit, developed with USAID and MIT D-Lab, provides concrete workshops and guidance for organizations across the sector on operationalising principles like fairness.
What’s worth noting is how they communicate about it. The Humanitarian AI Code of Conduct, developed collaboratively with NetHope members, is explicitly framed as a sector-wide effort to “demystify AI technology and set responsible standards before the use of such technology becomes ubiquitous.” That framing matters. It names the problem directly — AI is arriving fast, the risks are real, and the sector needs to build standards now rather than retrofit them after something goes wrong. Daniela Weber, director of NetHope’s Center for the Digital Nonprofit, wrote plainly in 2024 that AI risks “further augmenting new digital gaps in skills, protection, inclusion, transformation” and that “as humanitarians, we need to address and mitigate those risks for the sake of the communities we serve.” No euphemisms. No “exciting opportunities alongside challenges.” Just: here are the risks, here is the work.
That’s not a minor communication choice. In a sector where funders, partners, and beneficiaries are all evaluating whether to trust you, naming risks publicly is a form of credibility-building that polished comms can’t replicate.
Oxfam’s Rights-Based Governance Model
Oxfam International’s January 2025 submission to the UN Working Group on Business and Human Rights grounds its AI governance in the UN Guiding Principles — not in vague commitments to “responsible use,” but in the specific framework that governs corporate accountability for human rights. That’s a significant positioning choice. It means Oxfam is holding itself to the same accountability standards it holds companies to in its advocacy work. The message to communities and donors is implicit but legible: we’re not applying different standards to ourselves.
Candid’s Approach to Honest AI Disclosure
Candid, the nonprofit sector’s main research and data organization, launched a LinkedIn newsletter using AI to draft content from recycled thought leadership pieces — and was transparent about how. That’s a small example, but it gestures at something real: normalising disclosure rather than hoping nobody notices. Their published guidance on responsible AI use policies explicitly calls out the requirement to “notify donors that you’ll be using data you collect from your interactions to train an AI algorithm” and to provide an opt-out period for existing donors. The default, in other words, shouldn’t be silence. The default should be telling people.
Universities That Stopped Using Broken Tools
The universities that disabled Turnitin’s AI detection are worth dwelling on because they did something genuinely difficult: reversed course, publicly, on a tool they’d already adopted. That’s hard for institutions. It requires acknowledging error, communicating the change to students and faculty, and accepting that some people will use the reversal to question the original decision. Most institutions prefer to quietly walk things back. The ones that named the problem — bias against non-native English speakers, false accusation rates between 10% and 20% in diverse classrooms — gave students something valuable: an explanation. Not a gesture, an explanation.
The University of North Carolina’s approach is instructive in a different way. UNC implemented a “dual responsibility model for transparency,” requiring faculty to disclose when they use AI to develop course content and students to acknowledge any AI assistance in completing assignments. It doesn’t pretend AI isn’t being used. It makes the use visible.
Why Transparent Communication Actually Works Here
There’s research on this, not just intuition. A study published in Humanities and Social Sciences Communications found that sharing information about how AI works — and acknowledging its limitations and efforts to overcome its imperfections — mitigates negative attitudes toward AI and builds trust not just in the technology but in the organization using it. The finding cuts against the instinct to manage perception by saying as little as possible. Acknowledging imperfection, it turns out, builds more reputational trust than projecting confidence.
For NGOs and nonprofits, this matters because the trust relationship is the whole game. Public concern about AI has held relatively steady since the release of ChatGPT in late 2022. Donors and communities haven’t been reassured into comfort by AI companies’ communications. They’ve watched the headlines. The organizations that are proactively telling them how AI is being used — what data is collected, what decisions it informs, what it can’t do — are the ones differentiating themselves in a landscape where trust is genuinely scarce.
Only 15% of surveyed US nonprofits have a responsible AI use policy. Which means the organizations that do have one, and communicate it clearly, are already ahead. Not because a written policy is magic, but because having one is evidence of having done the thinking. Donors, grant makers, and community members can see the difference between an organization that has worked through its AI use carefully and one that’s just hoping nothing goes wrong.
What This Actually Requires
Honest communication about AI isn’t primarily a writing challenge. It’s an organizational one.
It requires leadership that’s willing to say “we tried this tool and it didn’t work” rather than defaulting to silence. It requires involving staff, community members, and beneficiaries in policy conversations rather than presenting finished documents for sign-off. Students, for instance, call for clear and comprehensive guidelines on acceptable AI use — but the literature also underscores the need for genuine dialogue with students to reconcile differing perspectives, not just top-down communication. That’s a harder process than drafting a policy. It’s also the one that produces buy-in.
It requires being specific. “We use AI responsibly” is not communication. Responsible communication means telling donors what data is being used, how AI informs decisions, what it can’t do, and how to opt out. It means explaining to a community what a predictive model flags and what happens as a result. Vagueness reads, correctly, as evasion.
And it requires honesty about the unresolved parts. The 2025 AI Equity Project found that 76% of nonprofit organizations lack any AI strategy at all. Most organizations using AI right now are figuring it out as they go. Saying that — “here’s what we’re using, here’s what we’re still working out, here’s who to contact with concerns” — is a more honest and more trust-building statement than a polished AI ethics page that suggests everything is under control.
The Broader Stakes
There’s a cynical version of this argument: organizations are being transparent about AI because it’s good for their brand. Maybe. But the people actually doing this work — the NetHope AI ethics team, the faculty learning communities at UNC, the university administrators who pulled an AI tool after seeing the bias data — mostly seem to be doing it because they think it’s right, and because they understand who pays the price when it goes wrong. It’s not the organization. It’s the student falsely accused of plagiarism. It’s the aid recipient whose data gets misused. It’s the donor who finds out their giving history was fed into a model they never consented to.
The organizations winning on communication aren’t the ones with the best AI messaging. They’re the ones that decided the question “what should we tell people about how we use AI?” deserves the same rigour as the question “what AI should we use?” That’s a harder standard to meet. It’s also the one that earns the trust that mission-driven work depends on.
Sources and further reading: NetHope AI Ethics Toolkit · NetHope Humanitarian AI Code of Conduct · Candid: Getting Started on a Responsible AI Use Policy · GlobalGiving: Responsible AI Use Policies · Whole Whale: Top Nonprofit AI Policies 2025 · AI in Education Ethics (AI-Tutor) · High-Risk by Algorithm: AI Bias Threatens Student Equity · Nonprofit Leadership Alliance: Using AI, Maintaining Trust · The Nonprofit Alliance on AI · AI Transparency as Trust-Building (Nature/Humanities & Social Sciences Communications)
