An AI system I built was used in a way I never intended.

I designed a productivity analysis tool for a manufacturing client. It identified bottlenecks in production workflows. Good use case. Clear value.

Six months later, I discovered they'd also started using it to rank individual workers by "efficiency" and making employment decisions based on the rankings.

The model wasn't designed for this. The training data wasn't suitable for this. The evaluation metrics weren't appropriate for this. But it was technically possible, so someone did it.

This experience changed how I approach AI ethics. It's not just about building fair models. It's about thinking through how your system COULD be misused and building in guardrails.

What I do now: every design document includes a "potential misuse" section. Output is constrained to the intended use case — if the tool is for workflow analysis, individual-level scores aren't produced. Clear documentation states what the system is designed for and explicitly what it's NOT designed for.

You can't prevent all misuse. But you can make it harder, and you can make your intentions clear.

The question isn't just "does this model work?" It's "what happens when someone uses it for something we didn't intend?"