Trust, but Verify: Using AI Wisely In a Nonprofit World
I have spent some time the past few weeks in seminars and discussions about artificial intelligence, especially how nonprofits are using it. One theme keeps coming up. Funders and donors increasingly want their dollars to go to programs rather than to paying people for nonprofit labor, so many organizations are turning to AI to supplement the work.
For a lot of them, that is not a trend. It is a matter of survival, and sometimes the choice is between using AI to keep serving the people they support or cutting back because they don’t have the staff or volunteers to do the work.
I want to say clearly that I am not here to argue against AI. Most of us have been using it for years. Every time we ask Google a question or ask Alexa to play a song, we are relying on it. So I understand why nonprofits are leaning in, and the numbers show the sector is moving fast. One 2026 survey of 917 nonprofit and NGO respondents found that 98 percent were using AI in some capacity.
That brings me to the question I have been sitting with: are we pairing that speed with oversight?
Governance is catching up, and that is a good conversation to have
The same research suggests that adoption is moving faster than governance. In a December 2025 benchmark of 346 nonprofits, 47 percent reported having no AI governance policy. A September 2026 survey from NTEN and Bridgespan found that fewer than 40 percent of groups had written policies on safe and responsible AI use. Bridgespan’s own release on the research notes that human oversight and privacy practices are part of what supports responsible use.
I see this as an opportunity. Most of us are still writing the playbook, which means we have room to build good habits early.
Think of AI as a new team member
Here is the way I think about it. Imagine you hired a new staff member and, from their very first day, you never reviewed their work. Nothing was checked for completeness or accuracy before it went out to the public or to a funder. Most of us would never do that, however talented the person was.
AI deserves the same care. It can draft, summarize and research at remarkable speed, but human oversight and discernment are what make sure the information is real before we put our organization’s name on it.
What does it mean to “poison the well”?
One idea that came up in a conversation this week in a seminar on evaluation best practices shared by The Nonprofit Association of Washington has stayed with me. Some people are now deliberately posting misinformation across many websites, hoping AI tools will pick it up and repeat it as fact. It is often called data poisoning, or poisoning the well.
As it was explained to me, I could post in enough places that Trudy Soucoup is the Prime Minister of Canada, and an AI tool gathering information from the internet might eventually report it as true. I am not, in case you were wondering.
I wanted to know whether this is a real concern, so I went looking. Researchers at Anthropic, the UK AI Security Institute and the Alan Turing Institute found that just 250 malicious documents could create a backdoor vulnerability in language models ranging from 600 million to 13 billion parameters. Size did not seem to protect the larger models. Those documents were a tiny fraction of the training data, about 0.00016 percent.
To be fair, the researchers tested a narrow, low-risk behavior, and they shared their work so that defenders could prepare. It also studied what happens when a model is trained, which is a little different from an AI tool searching the web for you. Still, it raises a fair question for all of us. If a small number of planted documents can influence a model, how confident are we that every answer we receive is accurate?
A few questions worth asking in your organization
Who reviews AI-generated content before it is shared with donors, funders or the public?
Do we have a written policy that says how and when we use AI tools?
How will we verify facts, figures and sources before we rely on them?
What information should never be entered into an AI tool?
Be careful out there
AI may well help many of us keep our doors open and our missions moving, and I am glad to see organizations find creative ways to stretch limited resources. My encouragement is simple. Keep a human in the loop, check what you scrape, and stay curious. It really is a brave new world, and with good judgment and a little healthy skepticism, we can navigate it together.
What is your organization doing to review its AI-generated work? I would love to hear your approach.
Sources
Comments