By the Shadow AI Policy team
**Nonprofits are using AI tools constantly — for grant writing, donor outreach, volunteer coordination, and communications — and most of them have no policy governing any of it.** This post covers the five governance areas that matter most for nonprofit organizations: protecting donor PII when staff use AI tools, disclosure obligations in AI-assisted grant writing, how boards should own AI oversight, scoping AI use for volunteers and contractors, and what free-tier AI tools actually do with your data.The biggest risk for nonprofits isn't using AI — it's using AI tools that treat your data as training input without anyone at your organization knowing it. Before you write a single policy rule, find out which free tools your staff are already using and read their default data retention terms. That single step will tell you what you're actually governing.
Donor records — names, giving history, contact information, employer data, wealth screening results — are among the most sensitive data a nonprofit holds. When a staff member pastes a donor list into a free AI chatbot to draft a segmented appeal, that data has left your environment. Whether it stays there depends entirely on the tool's data policy, not your intentions.
The practical rule is simple: no donor PII goes into any AI tool that hasn't been reviewed and approved by your operations or legal lead. That includes full names paired with giving amounts, email addresses, employer affiliations, and any wealth screening data your CRM contains. If a staff member needs AI help drafting a donor communication, they should draft with anonymized placeholders ("Donor X, $5,000 annual gift") and substitute real data only inside your CRM or email platform — never inside a public-facing AI interface.
For nonprofits subject to state privacy laws — California's CCPA (Cal. Civ. Code § 1798.100 et seq.) applies to many nonprofits that meet the revenue or data volume thresholds — sharing donor data with an AI vendor without a proper data processing agreement may create disclosure obligations or liability. Even if your organization falls below CCPA thresholds, donor trust is a direct funding risk. A data exposure event involving donor records is harder to recover from than almost any other operational failure.
Build a short approved-tools list. It doesn't need to be long — even three or four tools your staff can use for donor-facing work, with documented data handling terms, is enough to move from ungoverned to governed. Post it somewhere visible — your intranet, a shared drive, a Slack channel — so staff don't have to guess.
AI-assisted grant writing is now standard practice, but grant funders are increasingly asking about it. Some major foundations have added language to their RFPs requiring disclosure of AI use in applications. Others have not — yet. The risk of staying silent is that your organization's position becomes reactive: you disclose only when caught, rather than as a deliberate practice.
The practical governance move here is a standing internal rule: any grant application that used AI assistance — drafting, editing, data summarization, narrative structuring — gets a one-sentence disclosure in the cover letter or application narrative. Something like: "Portions of this application were drafted with AI writing assistance and reviewed and edited by [Name], [Title]." This is not a legal requirement in most cases. It's a trust and reputation decision, and it protects the organization if a funder later asks.
The harder question is content accuracy. AI tools can generate plausible-sounding program statistics, outcome data, or research citations that are wrong. Your policy should require that any AI-generated content in a grant application be fact-checked against primary sources before submission. Assign that responsibility to a named person — not "the team." Grant applications with fabricated data — even inadvertently — can result in clawbacks, funder relationship damage, and in cases involving federal grants, potential False Claims Act exposure under 31 U.S.C. § 3729.
Most nonprofit boards have added cybersecurity and data privacy to their risk oversight responsibilities in recent years. AI governance belongs in the same bucket, but it rarely gets there unless someone puts it on the agenda explicitly.
Your board doesn't need to approve every AI tool your staff uses. What it does need is:
If your organization has a Finance and Audit Committee, AI data governance fits naturally there. If you have a separate Risk Committee, even better. What you want to avoid is a situation where no board-level body has visibility into AI use — because if something goes wrong with donor data, "we didn't have a policy" is not a defensible position for a board member with fiduciary obligations.
The ask of the board is minimal: review and ratify the policy once a year, receive a brief staff report on any new tools or incidents, and make sure the executive director has the authority to enforce the rules. That's a 20-minute agenda item, not a new committee.
Volunteers and contractors are the most common source of ungoverned AI use at nonprofits — not because they're careless, but because they typically aren't onboarded with the same rigor as staff. A communications volunteer helping draft newsletters, a contract grant writer, a pro-bono web developer: all of them may use AI tools as a matter of course in their own work, and unless you've told them otherwise, they'll assume that's fine.
Your AI acceptable use policy should explicitly cover non-employees who handle organizational data. That means:
For longer-term contractors — grant writers, marketing consultants, IT support — ask them what AI tools they use professionally and whether those tools' terms permit processing client data. A contractor using a business-tier AI subscription with a proper data processing agreement is a different risk profile than one using a free consumer tool. Shadow AI — tools used outside IT visibility — is especially common in contractor workflows because contractors bring their own tool stacks.
Most nonprofits can't afford enterprise AI licensing, which means staff gravitate toward free tiers of popular AI tools. The governance challenge is that free tiers often have materially different data terms than paid tiers. This isn't about the AI output quality — it's about what happens to the data you put in.
Here's a practical comparison of how data handling differs across tiers for common AI tools:
| Tool / Tier | Default Data Used for Training? | Data Processing Agreement Available? | Nonprofit Discount / Free Tier? |
|---|---|---|---|
| ChatGPT Free (OpenAI) | Yes, by default (opt-out available) | No (paid plans only) | No nonprofit tier; free tier only |
| ChatGPT Plus / Team (OpenAI) | Off by default on Team plan | Yes (Enterprise plan) | No; standard pricing applies |
| Microsoft Copilot (M365) | No (tenant data not used for training) | Yes (covered by M365 DPA) | Yes — via Microsoft Nonprofit program |
| Google Gemini (Workspace) | No (Workspace data not used for training) | Yes (covered by Google Workspace DPA) | Yes — Google for Nonprofits program |
| Claude Free (Anthropic) | May use conversations to improve models (see Anthropic Privacy Policy) | No (paid/API plans only) | No nonprofit tier currently |
Note: AI tool policies change frequently. Check each vendor's current terms before making access decisions. The above reflects publicly available terms as of mid-2025.
The practical takeaway: if your nonprofit already uses Microsoft 365 or Google Workspace, you likely have access to AI features under a data processing agreement at little or no additional cost through the respective nonprofit programs. That's a better starting point than a free consumer AI tool with no DPA, even if the consumer tool feels more capable for some tasks.
A data processing agreement doesn't make an AI tool safe — but the absence of one means you have no contractual basis for expecting your data to be protected. For donor data, that's not a theoretical risk.
For tasks that don't involve sensitive data — brainstorming program names, drafting social media copy with no donor references, summarizing publicly available research — free tools are fine. The policy decision isn't "ban free tools." It's "draw a clear line between tasks where free tools are acceptable and tasks where they aren't," and write that line down. A two-column list in your policy document (Approved Uses / Restricted Uses) accomplishes this without bureaucratic overhead.
If you're starting from scratch, an AI acceptable use policy template structured around data sensitivity tiers is the fastest path to a working policy. You can also generate a tailored policy kit that reflects your organization's specific data types and risk profile without starting from a blank page.
About Shadow AI Policy: We build AI acceptable use policy tools for HR and operations teams at 50–500 person companies. We publish guides on shadow AI, acceptable use policies, and AI governance, updated as regulations and AI tools change.
The biggest risk is staff or volunteers entering donor records — names, giving history, contact information, wealth data — into free AI tools that use conversation data to train their models. Unlike a corporate data breach, a nonprofit donor data exposure directly damages the trust relationship that fundraising depends on. Most free AI tools' default terms permit some form of data use; only paid business tiers with data processing agreements typically provide contractual data protection. The fix is a short approved-tools list tied to data type, not a blanket ban on AI.
There's no universal legal requirement to disclose AI use in grant applications, but a growing number of foundations are adding explicit AI disclosure requirements to their RFPs. Even where it's not required, proactive disclosure protects your organization's reputation if a funder later asks. The more pressing concern is accuracy: AI tools can generate plausible but incorrect program statistics or research citations, and submitting inaccurate data in a federal grant application can trigger False Claims Act liability under 31 U.S.C. § 3729. Assign a named staff member to fact-check all AI-generated content before submission.
Yes — and a small team is actually the easiest environment to implement one. A one-page document covering what data can't go into AI tools, which tools are approved, and who owns decisions about new tools is enough to move from ungoverned to governed. The risk doesn't scale with headcount: a two-person development team with access to your donor CRM has the same data exposure risk as a twenty-person team. A simple policy also helps with board governance obligations, which apply regardless of org size.
Add a one-paragraph AI use clause to your volunteer agreements that mirrors your staff policy: no donor PII, board materials, or unreleased financial data in AI tools, and disclosure if AI is used in deliverables. Brief volunteers on this at onboarding — a two-minute verbal explanation plus the written clause is enough. For contractors, require them to confirm in writing that any AI tools they used to produce deliverables did not process your organization's confidential data, and ask them to identify what tools they use professionally so you can assess the data terms.
Tailored to your industry and the AI tools your team uses. Free preview, $79 one-time or $149/mo with monthly updates.
Generate my policy kit →