Keep a human hand on the wheel.
AI Training & Governance
The Mundanehas a new trick. It offers your team a shiny genie that grants wishes — write this grant report, draft that email, summarize these notes. And the genie delivers, which is exactly why it's dangerous: staff start feeding it the org's secrets, and every so often it lies out loud, confidently and in your name. That's Chaos wearing a helpful smile — the hallucination, the leaked donor list, the embarrassing headline.
AI can be a genuine force-multiplier for a lean team. The trick is to take the wishes without getting cursed. The Crew does it by writing one short pact and keeping a human hand on the wheel. (Our founder co-authored the City of Long Beach's Generative AI Guidance — the same playbook, scaled to your size.)
AI is a tool, not an oracle. Use it boldly — but keep a human hand on the wheel and your values in the driver's seat. Don't let a robot speak for the people we serve.
I literally automate away busywork for a living, so trust me: never feed me secrets you wouldn't post publicly, and never ship my output unread. One page of rules, an approved-tools list, training-data settings off. That's governance. #AutomateThis!
The Crew's playbook: a one-page AI use policy
You don't need a 40-page framework. You need one page your whole team will actually read. Start with these five rules:
- 1
Never paste sensitive data into consumer AI tools
No donor records, client PII, health info, or anything confidential into free public chatbots. Treat the chat box like a postcard — assume someone could read it.
- 2
A human reviews anything that goes public
AI drafts; people approve. Nothing AI-written reaches a donor, grantor, or the public without a human checking it first.
- 3
Keep an approved-tools list
Name the specific AI tools staff may use, and the settings to use them with. "Use AI" is not a policy; a short list is.
- 4
Disclose AI-generated content where it matters
Be honest when content is AI-assisted, especially imagery and anything attributed to a person or beneficiary.
- 5
Protect your org's data settings
Turn off training/data-retention where the vendor allows it, and prefer tools that offer a business tier with a data processing agreement.
The AI Use Register worksheet
The policy becomes real when you can see, on one page, every place the Whispering Genie touches your org. A five-column table, revisited every quarter — so no use slips into the shadows without a human watching:
| AI tool / use | What it does | Data it touches | Human check | Approved by |
|---|---|---|---|---|
| ChatGPT — drafts | First drafts of emails & posts | Public info only; no PII | Staff edits & fact-checks before sending | Comms lead |
| Meeting transcription | Notes & summaries from calls | Internal staff voices; sometimes client names | Reviewer redacts names before sharing | Ops lead |
| Resume screening | Sorts applicants for a role | Applicant PII | Human reviews every shortlist & rejection | HR / hiring manager |
| Customer chatbot | Answers FAQs on the site | Visitor questions; no account data | Weekly transcript review; human handoff path | Program director |
| Image generation | Illustrations for posts & decks | Prompts only; no real people's likenesses | Disclosed as AI-made; checked for bias | Comms lead |
Add a row for every place AI shows up — new tools, new uses, the “just this once” experiments. The version of this table with a real human in every “Human check” cell is the version that keeps the genie granting wishes instead of leaking secrets.
Avoiding embarrassment
AI fails in specific, predictable ways. Name them so your team can catch them:
- Hallucinations — it invents facts, names, and citations that sound right. Verify anything factual before it's public.
- Bias — it can reflect stereotypes. For mission-driven work, read with a critical eye and a diversity lens.
- Voice drift — generic AI text can flatten your org's warmth. Edit it back into your own voice.
- Speaking for others — never let AI put words in the mouths of beneficiaries or generate their likeness without consent.
Training your team
Policy without literacy is just a poster. Spend an hour helping staff understand what AI is good at, where it fails, and how to use the approved tools well. A free place to start: our LearnAI course — a no-cost, no-login interactive AI literacy course with a voice-powered guide.
How Chaos & The Mundane win
- No policy at all — every staffer improvising their own rules.
- Banning AI outright, which just drives it into the shadows on personal accounts.
- Pasting donor data, client info, or case notes into free public chatbots.
- Publishing AI output nobody fact-checked — and inheriting its mistakes.
Common questions
▸ Should we just ban AI to be safe?
No — banning AI usually backfires. Staff use it anyway on personal accounts, where you have zero visibility or controls ("shadow AI"). A short, clear use policy plus an approved tool is far safer than a ban nobody follows.
▸ What's the one rule that prevents most AI data leaks?
Never paste confidential or personal data — donor records, client info, health details — into consumer AI tools. Many free tools may use your inputs to train their models. If it shouldn't go on a postcard, it shouldn't go in the chat box.
▸ How do we stop AI from embarrassing us?
Keep a human in the loop. AI can hallucinate (state false things confidently) and carry bias. Fact-check anything public, watch for biased or off-brand output, and never let AI speak on behalf of the people you serve without their consent.
▸ Do we really need a written AI policy if we're small?
Yes, and it can fit on one page. A small team is exactly where one informal decision becomes everyone's habit. A one-page policy — what's allowed, what's never allowed, and which tools to use — takes an afternoon and prevents most problems.
The next leg of the journey
Prefer a story? See these ideas play out in our comic-book field guide, The Whispering Bargain.
Ready to reach your Pitch?
You don't need a full-time CTO to do this right. Island Pitch works as your Fractional CTO — senior technology leadership at a nonprofit's budget, helping you choose well, lock the doors, and sleep at night.
Get the whole field manual
The full Right-Way Tech Guide — all seven chapters, including a one-page AI use policy template.
The field manual is yours — grab it now.
Our signup form is being upgraded. In the meantime, here's the full PDF — no form to fill out.
Download the field manual (PDF)Want occasional tech tips by email too? Talk to a Fractional CTO above — we'll get you on the list.