A retrospective on AlignIQ’s ethical AI course: the wins, the surprises, and the upcoming adaptation
With NPB’s Spring Training Program, we set out to teach ethical AI. We ended up learning quite a bit ourselves!
Over two cohorts of nonprofit practitioners, AlignIQ ran a course designed to help mission-driven mid- or senior-level staff identify one real workflow, experiment with AI thoughtfully, and leave with something they could actually use (and not just a certificate). Here’s what we saw.
Small wins are big indeed.
The course was best when someone had this moment: “Oh, that’s how this works.” And at times, students became the masters. One participant built a Python-based data sanitization tool after our session on data safety so that they could process documents with personal information without pasting raw personal identifying information (PII) into an AI model. Another sanitized spreadsheet data manually and reported being, in their words, “miles ahead,” likely saving weeks of work. A third discovered that customizing an LLM with key information upfront transformed their outputs for the better.
Not every win was about speed or quality. Our post-survey responses indicated that most participants were now using multiple verification checks. Some even said their AI-augmented workflow took about the same time once verification was included. Is this really a failure though? This outcome is measuring success correctly: building capacity to determine where processes actually benefit from AI, rather than treating speed and efficiency as the ends-all-be-all.
These ideas weren’t fluky transformations. They were specific, doable, customized, and repeatable.
Data safety isn’t a side door… It’s the whole house.
In both cohorts, the end-of-course survey surfaced the same blocker: data security. Even after we adjusted the second cohort to lead with safer use cases and earlier safety framing, participants were still wrestling with it at the end.
That’s actually good news. It means participants were taking their work responsibilities seriously, not just optimizing for speed. The challenge is that “what’s safe to put into AI” is genuinely hard to answer for a diverse group of orgs with different tools, different data, and often no internal policy to reference yet.
For organizations without clear AI guidance, the most empowering next step may not be “use AI on a painful workflow.” It may be “find a relevant use case that involves no sensitive data input at all.” Not so much a workaround as a sandbox on-ramp. Confidence is what makes the harder conversations possible later.
The questions participants asked were concrete and sophisticated. One questioned the paid version of) Claude’s zero data retention policy and whether it was still safe to share meeting notes or strategy. Another asked whether Gemini adds risk if their organization is already using Google Workspace. The answers to these are… maybe? We’re happy to help you talk this through!
Another raised something that made us really stop and consider: AI is already built into tools their partners use sometimes unavoidably. This is happening right now, not hypothetically. These procurement and privacy questions that org leaders are sitting with right now formed the most uncertain part of the learning, but it’s one we necessarily learned from.
Peer exchange was the real engine.
Chat participation mattered more than we expected. Useful comments in the chat often became subtle confidence-building moments for the whole group, not just for the person asking. Breakout rooms worked similarly: participants generally liked them as a space to discuss practicalities, and after we adjusted to fewer but longer sessions, the discussion was usually active and generative. Our topic turned out to be well-suited for peer learning, because many attendees had already started experimenting with AI prompting and had real field observations to share. That peer knowledge was at least as valuable as anything we put in the slide canvas.
Participants showed up braver than the moment required.
No one in the world feels fully prepared for AI right now. And yet, cohort after cohort, participants engaged with hard questions, challenged AI outputs, flagged hallucinations, and pushed back when something felt off. In one hallucination exercise, participants caught numbers that seemed too precise, legal details that “felt off,” possible misattribution, and subtle bias around women and candidates of color. Participants noticed when custom instructions made outputs worse, not just better. That’s exactly the judgment muscle that makes AI use sustainable rather than reckless. The students really went from “beginner” to “sophisticate” in many cases.
One outcome we hadn’t fully anticipated: participants left with material they could bring back to their organizations. The eco-friendly policy levers were particularly well received (if you email us, we can share!); people could present or adapt it internally, which meant the course wasn’t just personal upskilling. It really became a learning worksheet monster that couldn’t be tamed! That’s the kind of ripple effect that doesn’t show up in a satisfaction survey but matters quite a lot.
What’s next?
We’re repackaging what we built. Anchor workflows, decision trees, data safety triage, verification habits, and peer exchange all translate well beyond the nonprofit cohort context. We’re especially interested in what it looks like for individuals and small teams who are navigating AI adoption without the safety net of an organizational policy to lean on. Sound like you? Get on the list for what’s coming.
In the meantime, if your organization is thinking about AI governance, data safety, or how to build staff capacity without the chaos, let’s talk.
- On data safety paralysis: Wanted: Nonprofit Staff Who Can Wrangle AI and Still Care About People
- On building another course (abortively): AI Made Me Question Everything, So I Built a Course on AI Safety
- On delegating to AI thoughtfully: Delegating to AI: A Small Experiment in Letting Go
[In this post, we used AI for polish, not purpose]
Leave a Reply