Nobody Checks This: The Most Dangerous Phrase in AI Projects

The Most Dangerous Phrase in AI Projects

"Nobody checks this anyway."

I hear this phrase more often than I'd like. In review meetings. In deployment discussions. During security assessments. It sounds pragmatic. It is a governance failure.

Last week: An AI model was live in production, processing customer data – and nobody in the room could tell me exactly what data was flowing into it. Not the project manager. Not the developer. Not the data protection officer.

From Negligence to Governance Failure

The assumption "Nobody checks this" is not negligence. It is systemic governance failure. Negligence would be knowing something needs to be checked and not doing it. That would be fixable. What I see instead: no one ever clearly defined who is responsible for checking. So no one checks.

Three Questions Every AI Project Must Be Able to Answer

First: Which data flows into the model? What data is used for training, inference, and evaluation? Who classified and approved this data?

Second: Who validates the outputs before they influence decisions? AI systems deliver probabilities, not facts. The question of who validates outputs before business decisions must be answered – in writing, with names.

Third: What happens when the model is wrong, and who is responsible? A credit model that systematically makes wrong decisions. A supply chain model that misses bottlenecks. Who is accountable? Who notifies the customer? Who stops the system?

If None of These Questions Are Answered

Then the project isn't innovative. It's reckless. The solution is not a 50-page AI governance framework gathering dust in a drawer. It is a clear review regime: who checks what, when, with what outcome. Documented. Repeatable. Assigned to individuals, not just roles.