Where Your Donor Data Actually Goes
Associations hold donor data most companies would treat as radioactive: membership records, donor giving history, sometimes health or professional licensing information. It’s the kind of data that gets locked down in every other context — except, often, the one that matters most right now.
None of it is supposed to end up inside a third-party AI model’s training data. Most of it ends up there anyway, and not through malice. It’s a well-meaning staffer pasting a donor list into a chatbot to save twenty minutes writing a summary, without stopping to think about where that text actually goes once it’s typed in and submitted. The tool doesn’t ask permission. It just processes what it’s given.
The uncomfortable part is that this has almost certainly already happened at associations that don’t know it happened. Nobody’s watching for it, because nobody has written down what “watching for it” would even mean — no list of what actually counts as sensitive, no rule about which tools are approved for which kind of data, no one whose job it is to notice when the line gets crossed.
And “ending up in training data” isn’t the only version of this risk, just the least visible one. There’s also the more mundane version: a well-meaning staffer using a free-tier AI tool that stores conversation history indefinitely, or sharing an AI-generated summary externally without realizing the underlying data was never meant to leave internal systems in the first place. None of these staffers are doing anything malicious. They’re doing what any reasonably resourceful employee does when a tool makes a tedious task faster — using it, without a clear signal from leadership about where the boundary actually is.
How to Keep Donor Data Safe
Closing that gap takes specificity, not a general warning to “be careful with AI”:
- Name the sensitive categories explicitly. Not “personal data” in the abstract — donor giving history, licensing status, health information, anything the association would never want to see in a headline.
- Match tools to data types. Some AI tools are contractually safe for sensitive data; most consumer tools are not. The policy has to say which is which, tool by tool.
- Set a default for ambiguous cases. When staff aren’t sure whether something is sensitive, the default should be no — not “ask forgiveness later.”
- Build the check-in before the mistake, not after. A short, recurring review of what’s actually being pasted into AI tools catches this before a donor calls asking hard questions.
None of this needs to be invented from scratch. The National Institute of Standards and Technology publishes a free, voluntary AI Risk Management Framework built around four functions: govern, map, measure, and manage. It was written for organizations of every size, and its logic applies directly to donor data. Governing means deciding who is responsible. Mapping means knowing where sensitive data lives and which tools touch it. Measuring means checking whether the rules are actually followed. Managing means acting when they aren’t. An association does not need to adopt the full framework to benefit from it. Borrowing its structure is often enough to turn a vague concern into a short list of decisions a board can approve.
The goal isn’t to make staff afraid to use AI. It’s to make sure the sensitive data an association has spent years earning trust to hold doesn’t leave the building by accident, one convenience shortcut at a time. A board-approvable AI Governance Policy is where these lines get written down and made real — not as a warning, but as a working document staff can actually follow.
