TL;DR: A nonprofit can use AI with donor data safely when five controls are in place: AI accounts whose terms bar the vendor from training on the organization's data; data minimization, so the tool sees aggregated or de-identified data whenever that is enough; a firm rule that payment card and bank details never enter an AI tool; AI use that fits the privacy promises made to donors and the state privacy laws that apply; and a person who reviews anything AI-assisted before it reaches a donor. Set the controls before automating.
Development teams adopted AI early, for good reasons. Drafting appeals, segmenting lists, summarizing giving history before a visit, and writing acknowledgment letters at volume are all tasks AI handles well. Each of them, done carelessly, puts donor information somewhere the donor never agreed it would go.
In most nonprofits, staff are already using AI with donor information in some form. The leadership task is to set the conditions under which that use is safe and to make those conditions easy to follow.
What makes donor data different?
Donor data combines identity, money, and motive. A giving history shows what a person cares about, what they can afford, and often what they have lived through. Many donors give to causes connected to illness, addiction, faith, or family loss, so their giving can reveal far more than a mailing address.
Donors also extend trust on specific terms. The Donor Bill of Rights, published by the Association of Fundraising Professionals and adopted widely across the sector, includes the assurance that information about a donation is handled with respect and with confidentiality to the extent provided by law. Most nonprofits repeat a version of that promise in their own privacy policy. AI use has to stay inside it.
What are the real risks?
- Training and retention. Some AI tools, especially free consumer versions, may use inputs to improve their models or keep them longer than the organization would choose.
- Oversharing. A staff member pastes a full donor export into a tool when a summary or a few fields would have done the job.
- Access sprawl. Several people use several tools on personal accounts, and no one can say where donor data now lives.
- Errors that reach donors. An AI-drafted letter thanks a donor for the wrong gift, misstates a pledge, or references a sensitive circumstance.
- Legal exposure. A growing number of state privacy laws apply to nonprofits, and an AI vendor that receives personal data becomes part of the organization's compliance picture.
Which state privacy laws apply to nonprofits?
Many early state privacy laws exempted nonprofits. Several newer ones do not. Colorado and Oregon apply their comprehensive privacy laws to nonprofits, New Jersey's law reaches nonprofits that meet its processing threshold, and Delaware and Minnesota carve out only narrow nonprofit exemptions (Wagenmaker & Oberly, April 2025; see also Wiley).
Applicability depends on the states where donors live, the number of records processed, and the thresholds in each law, so counsel should confirm which laws apply. For AI purposes the practical work is the same in every case: know what personal data each AI tool receives, have a contract that governs it, and be able to honor a donor's request about their data.
How can a nonprofit use AI with donor data safely?
Five controls carry most of the weight.
1. Use accounts with the right terms
Use business or enterprise AI accounts managed by the organization, and read the data terms before approving a tool. Look for a commitment that the vendor will not train its models on your inputs, clear retention limits, and administrative control over who has access. Microsoft states that for Microsoft 365 Copilot, prompts, responses, and data accessed through Microsoft Graph are not used to train foundation models (Microsoft Learn). Confirm the current terms of any tool you approve, since providers revise them.
2. Give the tool the least data that does the job
Most development tasks do not need a donor's full record. Drafting an appeal needs the campaign story and a description of the segment. Analyzing retention needs counts and dates, not names. Preparing for a major donor visit may need a giving summary, which can be produced inside the donor database instead of exported. Default to aggregated or de-identified data, and treat named data as the exception that needs a reason.
3. Keep payment and bank details out entirely
Card numbers, bank account numbers, and routing details never enter an AI tool. Card data falls under the PCI Data Security Standard, and the organization's payment processor should be the only system that holds it.
4. Keep the promises made to donors
Read the published privacy policy and the language on giving forms. If either promises that donor information stays with the organization, sending it to an outside AI vendor has to fit that promise, usually through a contract that binds the vendor to act only on the organization's instructions. Update the privacy policy when AI use changes how donor data is processed.
5. Put a person between AI and the donor
Every message, letter, and report that reaches a donor gets a human read before it goes. The reviewer checks names, amounts, dates, and tone, and owns the result. This is the control that protects relationships, and it costs minutes.
Where does AI help most in development?
With those controls in place, the gains are real and measurable.
- Appeal and acknowledgment drafting from campaign briefs and segment descriptions, with gift details merged by the donor database itself.
- Visit preparation, using giving summaries produced inside the CRM.
- Grant writing from program data and outcomes, which rarely requires individual donor data.
- Retention and lapse analysis on aggregated or de-identified data.
- Board and finance reporting on giving trends, reconciled to the general ledger by the finance team.
Track the hours each workflow returns and report them with the rest of the development dashboard. AI that does not save meaningful time or improve accuracy should be retired.
What should the board ask?
Board members do not need to understand the tools. They need clear answers to a short set of questions, asked of staff once a year:
- Which AI tools touch donor or constituent data, and on what accounts?
- Do our contracts bar those vendors from training on our data?
- What donor data is never allowed into an AI tool, and how is that enforced?
- Which state privacy laws apply to us, and how does our AI use comply?
- Who reviews AI-assisted communications before they reach a donor?
An invitation
Novum's Technology & AI practice sets these controls with nonprofit leadership teams: the policy, the accounts, the vendor review, and the workflows worth automating. It works alongside our Finance & Accounting practice so giving data and the general ledger stay reconciled. If your development team is ready to use AI with confidence, book a Strategy Read or see how we advise, partner, and lead with nonprofits on technology.