Guide

Writing an AI-use policy for a nonprofit with paid staff

16 min read

By . Published . Last updated . Reviewed by Mavrin Labs on . How this guide was researched and checked.

A card reading "Writing an AI-use policy for a nonprofit with paid staff" beside the Mavrin Labs logo
The card that shows when this guide is shared or linked.

An AI-use policy is the short set of written rules your board adopts for how staff may use AI tools. It names the tools they may use, the data that stays out of them and the person who checks the work before it goes out. One role on your staff owns it.

The guide is written for the executive director or operations lead of a nonprofit with paid staff, a board and several funders. It covers what the policy should say, how to draft it with staff and how to take it to the board. It is not legal advice. Your board and your counsel decide what your policy says.

Sources were first checked Monday, October 19, 2026, and each citation gives the date it was last checked. Mavrin Labs reviews this guide every quarter, and each source's own page always wins over it.

What is an AI-use policy?

An AI-use policy is a short document the board adopts that sets which AI tools staff may use, which data stays out of them, who checks AI-assisted work and what the organization tells donors, so everyone handles AI the same way. One staff role owns it day to day.

It is not a list of products or a procedure manual, and it does not replace the privacy policy donors already read. It tells staff how the promises in that privacy policy apply once AI is involved, and the sections below set out what goes in it.

Why it matters: donors trust you with their data

A nonprofit needs an AI-use policy now because staff can reach AI tools on their own accounts long before the organization decides how those tools should be used. Donors, funders and the people you serve trust you with their information, and a written policy is how the organization keeps the promises it makes about that data.

The National Council of Nonprofits gives the clearest reason. It advises nonprofits to tell staff, donors, volunteers and the people they serve how their data is used, including whether it may go into AI tools.

It also advises a review of privacy policies every year (its article on data privacy, published Monday, January 22, 2024). A written policy is where those promises start.

The trust at stake is large. Giving USA 2026, the annual report on American philanthropy, puts charitable giving in 2025 at $617.20 billion, with $394.2 billion of it from individuals (the Giving USA announcement, checked Monday, November 16, 2026).

Most of those individual gifts leave a name, an address and a giving history in a donor database, which is exactly the data a policy keeps out of general AI tools.

What should an AI-use policy cover?

An AI-use policy should cover six parts: the tools and accounts staff may use, the data that never goes into them, the review before anything goes out, what you tell funders and donors, the records you keep, and who owns the policy. Each part fits in a few lines, and the outline further down gathers them on one page.

Tools and accountsThe policy owner approves each toolData kept outThe board adopts the ruleReview before sendingA named staff member checks eachdraftTelling funders and donorsThe board decides what to sayRecordsThe policy owner keeps themOwnership and yearly reviewThe policy owner reports, the boardconfirms
The six parts of the policy and who decides each. The filled parts are the board's decisions; counsel advises on both.

Keep the policy short enough that staff read it. Put the rules in the policy itself, and keep the list of approved tools in an appendix the owner can update without a board vote.

  • Which tools and accounts may staff use?

    Staff may use only the AI tools on the approved list, and only through accounts your organization controls. An organization account is one the nonprofit opens in its own name, under terms it has read and accepted. A personal account sits outside your control, so you cannot set or check what it does with your data.

    Providers often treat personal plans and business plans differently, and each plan's terms say whether the provider may train on what staff type. Read the terms of the exact plan each person uses, and switch off any setting that allows training on your data. The policy owner approves each new tool before anyone uses it for work.

  • Which data must never go into an AI tool?

    Donor records, client records and health information must never go into an AI tool unless that tool sits inside a system your organization approved for that data. Approval means someone has read the tool's terms for that kind of data and recorded the decision.

    Client records means anything about the people you serve, such as intake forms, case notes and eligibility details. Health information covers any note about a person's physical or mental health.

    The same rule should cover:

    • staff records, payroll and performance notes;
    • bank details, passwords and account numbers;
    • anything shared under a confidentiality agreement or a funder's data terms;
    • board discussions that are not yet public.

    When a task needs any of this data, the answer is a system built for it, with access limits and records, not a copy pasted into a general tool.

    Matching gifts to the books is one such task. The guide on how to reconcile donations against the books sets out a monthly routine that keeps donor records inside the donor database and the accounting system.

  • Who reviews AI output before it goes out?

    A named person on your staff reviews every piece of AI-assisted work before it reaches a donor, a funder, a client, the board or the public. NIST's generative AI profile of July 2024 lists confabulation, confidently stated false output, among 12 risks (NIST AI 600-1, checked Wednesday, November 18, 2026).

    Human review means that person reads the whole draft, checks names, figures, dates and tone against your own records, and answers for it once it is sent.

    AI tools can state wrong facts with confidence, so the review is a real check rather than a glance. The policy can let a message type go out without review only if the owner approves that type in writing and keeps a record of the decision.

  • What should you tell funders and donors?

    Funders and donors should hear how your organization uses AI in plain terms, whenever they ask and wherever your own promises say they will. Start with your donor privacy policy. If donor data may pass through an AI tool, the privacy policy is the place to say so.

    When a funder's terms ask how an application or report was prepared, answer exactly. Describe AI as a drafting aid only if that is how you use it, and never claim more for it than it does. Whether to say more than a funder asks is a decision for your board.

  • Which records should the policy keep?

    The policy should keep four kinds of records: the approved tool list with who approved each tool, the final version of anything sent, a log of incidents and questions, and the dates the board adopted and reviewed the policy. A final version counts as a record because of what it says and who received it, whoever or whatever drafted it.

    Fit AI work into the retention policy you already have, the written rule for how long you keep each kind of record and how you destroy it.

    The IRS instructions for Form 990 (the annual information return most tax-exempt organizations file) say an organization usually must keep records that support income, deductions or credits for at least three years from the date the return is due or filed, whichever is later (the instructions for Form 990, checked Monday, October 19, 2026).

    The same form asks whether the organization has a document retention and destruction policy. Other laws or your own policy may require longer.

  • Who owns the policy?

    One role on your staff owns the policy, most often the operations lead, and the board adopts it. The owner keeps the approved tool list, answers staff questions, logs incidents and brings the policy back to the board each year. Naming a role rather than a person keeps the policy working when people change jobs.

    Everyone the policy covers should learn it when they join, including volunteers and board members who use AI for your work. Make it easy to report a mistake, such as donor data pasted into the wrong tool, without blame, so the owner hears about it early.

How do you draft the policy with staff?

Draft the policy with the staff who will follow it, and start from how they already use AI rather than from a template. A policy written around the real tasks of each team gets read and kept. One written in the abstract gives staff no rule they can match to their own work.

Five steps keep the drafting short and practical, and each one involves the people who do the work every day.

  1. Ask what people use. Ask each team which AI tools they use, on which accounts and for which tasks, and make clear that nobody is in trouble for the answer.
  2. List the tasks. Sort the tasks into those that touch no protected data, those that touch it, and those that send something outside the organization.
  3. Form a small group. Bring together someone from operations, development, programs and finance, so each kind of data has a voice.
  4. Write it short. Keep the policy to one page, using the outline below, and move the tool list into an appendix.
  5. Test it on tasks. Run the draft against a few real tasks, fix what staff could not follow, and then send it to counsel.

How do you take the policy to the board?

Take the policy to the board as a short packet that asks for one decision: adopt the policy, with any changes the board makes. The board, or a committee it names, reads the draft, hears counsel's view and records the vote in its minutes. The policy then carries its adoption date and its next review date.

Put four things in the packet: the one-page policy, a summary of how staff use AI today, the approved tool list and the questions only the board can settle. Those questions are usually how far AI may touch donor communications and what the organization will tell funders and donors.

The IRS annual return for exempt organizations asks about several governance policies, though federal tax law generally does not require any particular one (the IRS instructions).

What does a one-page nonprofit AI policy template look like?

A one-page nonprofit AI policy template lists each part of the policy in the order staff will need it, with one or two plain sentences under each. Copy the table below into your own document, replace the wording with your organization's choices, and keep the list of approved tools as a separate appendix that the owner updates.

What does a one-page nonprofit AI policy template look like?
Part of the policy What it says
Purpose Why the organization uses AI tools, and the trust of donors and clients it protects
Who it covers Staff, volunteers, contractors and board members who use AI for the organization's work
Approved tools and accounts Only tools on the approved list, used through accounts the organization controls
Data kept out Donor, client, health, staff and financial records stay out of any tool not approved for them
Review before sending A named person checks every AI-assisted draft before it reaches anyone outside the staff
Telling funders and donors When and how the organization explains its use of AI, starting with the donor privacy policy
Records Final versions follow the retention policy, alongside the tool list and the incident log
Training and questions Everyone learns the policy on joining and knows whom to ask
Mistakes How to report data put in the wrong place, without blame, and who acts on it
Owner and review The role that owns the policy, the date the board adopted it and the next review date

Keep each line short enough to read in a minute. If a part needs more than a few lines, the detail probably belongs in a procedure the owner keeps, not in the policy the board adopts.

Which public framework can the policy draw on?

The AI Risk Management Framework, published by the National Institute of Standards and Technology (NIST), is a public framework a policy can draw on. NIST released it on Thursday, January 26, 2023 for voluntary use, and on Friday, July 26, 2024 it added a Generative AI Profile for generative AI, tools that write content from a prompt (the framework page).

The framework sorts the work into four functions: govern, map, measure and manage. Govern runs across the other three. It covers written policies, documented roles and policies for AI that comes from outside providers, and it calls for ongoing monitoring and a planned, periodic review (the framework's core).

A one-page policy is a small, practical piece of govern. NIST says it is revising the framework, so check its page for the current version.

The generative AI profile is the part most useful to a policy owner. It lists 12 risks that generative AI creates or makes worse, and data privacy, information security and confabulation (confident, made-up content) are among them (NIST AI 600-1, checked Wednesday, November 18, 2026).

Each one maps to a line of the policy: the data rule, the approved accounts and the review before sending.

How the guidance developed

Read in date order, the public guidance moved from general to specific in about eighteen months. NIST's framework of January 2023 set out AI risk practice for any organization (the framework page, checked Wednesday, November 18, 2026). The council's article of January 2024, cited above, applied nonprofits' data duties to AI tools by name.

NIST's generative AI profile of July 2024 then named the risks of tools that write, which is where a staff policy does most of its work. A policy built in that order starts from the framework's govern function, takes its donor promises from the council and its data and review rules from the profile.

How should the policy be reviewed each year?

Review the policy once a year at a set board meeting, and sooner when something changes. The owner brings a short report: which tools were added or dropped, what staff asked, any incidents, and whether the review before sending worked. The board then confirms the policy or adopts changes, and the minutes record the decision.

Kept as a few agreed figures, such as tools in use and incidents logged, that report can sit beside the board's other figures, the way a KPI dashboard for small business and nonprofit boards lays them out on one page.

Bring the policy back early when any of these happens:

  • staff want a new kind of tool, such as an AI agent, a tool that takes actions on its own rather than only drafting;
  • a provider changes the terms of a plan your staff use;
  • donor or client data reaches a tool it should not have;
  • a funder or donor asks how you use AI;
  • counsel tells you that a law affecting you has changed.

How can Mavrin Labs support the board's policy?

Mavrin Labs supports the board's policy by building systems that follow it. A system Mavrin Labs builds can include the approval steps, logs and records of what went out that a policy like this asks for. By default, nothing goes to a donor, funder or client until a person at your organization approves it.

Mavrin Labs never uses client data to train AI models, and your organization owns all of its data. Every engagement starts with a workflow assessment, and the written design plan for a build names the data the system will touch, so your policy owner can check it against the policy first.

Mavrin Labs gives no legal advice, and the policy stays your board's to write and adopt. The page on the data terms of every engagement sets out each term in full.

Key takeaways

  • Write the policy around the tasks staff already do with AI.
  • Keep donor, client and health data out of general AI tools.
  • Have a named person review every AI-assisted draft before it goes out.
  • Tell donors and funders how you use AI, in plain words.
  • Give one role ownership, let the board adopt the policy, and review it each year.

Sources behind this guide

Each source in this guide was last checked on the date given beside it, and the guide is reviewed quarterly. The table below pairs each outside fact with its publisher.

Sources behind this guide
Fact Source
Tell donors and others whether their data may be used with AI, and review policies each year National Council of Nonprofits article on data privacy
How long records are usually kept, and the return's question on a retention policy IRS instructions for the exempt organization return
A voluntary framework for AI risk, with a profile for generative AI NIST framework page
Charitable giving in 2025, and the share from individuals Giving USA announcement
The risks generative AI creates or makes worse NIST generative AI profile
Govern covers policies, roles, outside providers and periodic review NIST framework core

Corrections and changes to this guide, newest last:

  • Monday, November 16, 2026: the definition of an AI-use policy, the giving figures and the NIST risk list were added, each with its source.
  • Tuesday, November 17, 2026: the opening and closing sources lines were corrected to name every date a source was checked; the council's advice on data and NIST's note on confabulation were added; a paragraph pointing to a demonstration no longer published was removed; and the title was changed to the one the owner approved.
  • Wednesday, November 18, 2026: the NIST generative AI profile was opened again and now carries one checked date in both places it is cited; the policy diagram and the dated account of the guidance were added; repeats of the council's advice and of the counsel caution were cut, and two remarks that had no source were reworded.

Where to go next

The page for nonprofits shows the workflows Mavrin Labs looks at first for nonprofits with paid staff, and AI and your data sets out how a Mavrin Labs system handles your data. Workflow automation explains where a person approves each step.

Once the policy is in place, the six-question project finder suggests which workflow to try first, based on how your team spends its week.

If you are weighing outside funding for the work, the guide to funding AI and automation lists the programs and their rules. To talk through one workflow and the policy around it, start a conversation.

Frequently asked questions

Can staff use a personal AI account for work?

Staff should not use a personal AI account for work that touches your organization's data, and your policy can say so plainly. A personal account runs under terms the staff member accepted, not terms your organization chose, so nobody can check what happens to the data. Give staff an organization account for approved work instead.

Do we have to tell donors that we use AI?

Whether you must tell donors that you use AI is a question for your counsel, and what you choose to tell them is a decision for your board. The National Council of Nonprofits advises telling donors whether their data may be used with AI tools. Your donor privacy policy is the natural place to say it.

Should the board or the executive director approve the policy?

The board should adopt the policy, because it sets how the organization treats donor and client data, and that is a matter of trust the board answers for. The executive director or operations lead drafts it, owns it day to day and approves new tools under it. Counsel reads the draft before the vote.

Is this guide legal advice?

Nothing in this guide is legal advice, and Mavrin Labs gives none. The guide explains in plain words what a few public sources say, and it links each one so you can read the source yourself. Your board and your counsel decide what your policy says and how any law applies to your organization.

The Mavrin Labs mark

Who writes these guides

Client work at Mavrin Labs is delivered by a team in mixed roles, led by our founder. AI tools help with research and drafting, and a person at Mavrin Labs checks each fact against its named source before a guide is published.

Enquiry form

Start with one workflow

Tell Mavrin Labs what is stuck and which systems it touches. A person at the firm reads every enquiry and replies by email.

Start a conversation

Fields marked with an asterisk are the ones a reply needs.

So a reply can address you properly.

The address the reply goes to.

Optional. Give one only if you would rather talk than write; the reply goes by email either way.

Optional, and useful when it says what the work has to fit around.

The result you want, and when you would want it working. Two or three sentences are plenty.

The task as it actually happens, roughly how often it runs, and which systems it touches. Two links at most, please.

Optional.

What you send goes to the Mavrin Labs inbox and is used to answer you, and a phone number, if you give one, is used for the same thing and nothing else.

A person reads it and replies by email.

If the form does not send, write to info@mavrinlabs.com instead; it reaches the same inbox.

Which hours would you take back first?

Name the one workflow that eats the most of your week. Mavrin Labs reads it, and the reply comes from the people who would build the fix.

Start a conversation