How to Approve Free AI Tools Without Giving Away Donor Data

An AI tool approval process lets a nonprofit say yes to useful AI quickly while keeping donor and client data in accounts the organization controls. This guide covers the request form, the vendor terms check, the data tiers, who approves and how to catch tools nobody asked about.

Why does a free AI tool put donor data at risk?

A free AI tool is usually a personal account on consumer terms, so the organization does not control what happens to anything typed or uploaded into it. Once a donor list, a major gift memo or a client case note goes in, you cannot reliably pull it back out.

Consumer AI plans commonly keep conversation history, may let the vendor use content to improve its models unless the user changes a setting, and belong to whoever signed up. When that staff member or volunteer leaves, the history and any shared links leave with them. None of this shows up in your donor database, so nobody notices until a donor asks how their information was used.

The legal exposure is real for California nonprofits. The state's reasonable security statute requires a business that owns, licenses or maintains personal information about California residents to maintain reasonable security for it, and the definition of business in that part of the Civil Code expressly includes organizations that do not operate for profit. That law covers a person's name combined with specific data, such as Social Security numbers, financial account and card numbers with their access codes, and medical and health insurance information. Social service agencies hold a lot of it.

The broader California Consumer Privacy Act is different. As of October 2026, its definition of a covered business centers on entities operated for profit, so most charities fall outside it. Your own privacy policy and donor promises still apply, and those are what a board, a funder or a reporter will measure you against.

What should an AI tool request capture?

An AI tool request should capture who wants the tool, what job it will do and which data it will touch. Five questions on one short form are enough for most nonprofits.

• What is the tool, and what is the link to its pricing and terms?
• Who will use it: staff, volunteers, board members or all three?
• What task will it do, such as drafting appeals, summarizing meeting notes or analyzing a spreadsheet?
• What data will go into it: public content, internal documents, donor records, or client and program records?
• Does it connect to email, file storage or the donor database, and if so, what access does it ask for?

The last question matters more than it looks. Many AI add-ons ask for permission to read an entire mailbox or drive when someone clicks "Sign in with Google" or "Sign in with Microsoft." That one click can hand a vendor far more data than the person meant to share.

What should you check in the vendor's terms?

Check five things in the vendor's terms before anyone uses the tool with organization data: model training, retention, account control, sharing and integrations. If the free plan fails on training or account control, look for a business or nonprofit plan that passes. Think of it as a lightweight AI vendor risk assessment sized for a small team.

• Model training. Do the terms say your inputs and uploads are excluded from model training by default on the plan you will use? A setting that each user must remember to switch off is weak protection.
• Retention and deletion. How long does the vendor keep prompts, files and outputs, and can an administrator delete them?
• Account control. Can people sign in with the organization's Microsoft or Google account, with multifactor authentication, and can an administrator remove a departing user?
• Sharing. Can users create public links to conversations or files, and can an administrator turn that off?
• Integrations and subprocessors. Which other companies receive your data, and what exactly does each connector read?

Write down the plan name and the date you checked. Terms change, and the register you keep later depends on this record.

Which data should never go into an approved tool?

Even an approved tool needs a short list of data that never goes in. Three tiers cover most nonprofits, and they belong at the center of any nonprofit AI policy.

• Never, in any AI tool: Social Security numbers, bank account and card numbers, passwords, and client case notes, health or benefits information.
• Approved organization accounts only: donor names and giving history, grant drafts, board materials and internal memos.
• Any tool: content you have already published, such as your website, annual report and public program descriptions.

Card numbers deserve a specific note. If donation card data ever reaches staff by email or phone, keeping it out of AI tools is part of keeping it inside your PCI DSS scope. Put the three tiers on one page, and repeat them on the approval notice for each tool.

Who should run the AI tool approval process, and how fast?

One named approver should decide each request within a set number of days, with your IT partner checking the terms. A slow process pushes people back to free accounts, so speed is part of the control.

For most nonprofits the approver is the operations or finance director, with the executive director as the escalation point. Give three possible answers: approved, approved with conditions (for example, organization sign-in only and no donor records), or declined with a named alternative that already exists. A decline with no alternative is how shadow AI starts.

Record every decision in a simple register: tool, plan, owner, data tier allowed, date approved and next review date. That register lines up with the software and service provider inventories in CIS Controls Safeguards 2.1 and 15.1, and the CIS AI and LLM companion guide applies the same safeguards to AI. The NIST Generative AI Profile also treats third-party AI tools and their terms as a governance risk to manage. Review each approved tool once a year, or sooner when the vendor changes its plan or terms.

How do you catch tools nobody asked about?

You catch unrequested AI tools from sign-in and consent data, then route each one into the same approval process. Most organizations already have the data in Microsoft 365 or Google Workspace.

• Turn on alerts when a user grants a new third-party app access to organization email or files.
• Require administrator approval before users can consent to new apps that request broad permissions.
• Use web filtering to block or warn on AI sites that are not on the approved list.
• Review the list of connected apps monthly, and send each new AI tool its own request form.

For a walkthrough of that discovery step, see our guide to finding the AI apps your staff already use.

What should a nonprofit do this quarter?

Start with the form, the data tiers and the register, then add the technical controls. Most of it can be done in a few weeks.

If you are still choosing which tools to offer, see our guide to giving staff and volunteers AI on a nonprofit budget.

1. Publish the five-question request form and name the approver.
2. Publish the three data tiers on one page and add them to staff and volunteer onboarding.
3. List the AI tools already in use and run each one through the vendor terms check.
4. Move approved tools to organization accounts with multifactor authentication.
5. Turn on new-app consent alerts and require administrator approval for broad permissions.
6. Put the register on the agenda for an annual review with your board or finance committee.

How can s90 help with AI tool approvals?

s90 runs AI guardrails, new-app alerts and the approval register for nonprofits as part of an 18-month plan, so staff get useful AI tools quickly and donor and client data stays in accounts the organization controls. If you want a second look at the tools your team already uses, we are happy to start there.