What AI's hallucinations reveal about leaders

OpenAI recently explained why AI "hallucinates," because it's rewarded for confidence over truth. Many leaders are trained the same way, until they realize strategic honesty works better.

AI doesn't lie on purpose. It hallucinates because of how it's graded: a correct answer is rewarded. A wrong answer with confidence is sometimes still rewarded. "I don't know" earns zero reward.

So the model learns to always generate something, even when uncertain.

Leaders often fall into the same trap. They assume authority is about always having the answer.

Trap 1: competence = knowing the answer. Reality: competence is knowing how to find the best answer.

Trap 2: bluffing beats admitting not knowing. Truth: similar to how hallucination prevents AI adoption, once caught bluffing, a leader's credibility deteriorates fast.

Trap 3: decisiveness = certainty. Reed Hastings at Netflix made bold bets with incomplete data. But he was explicit about uncertainty: "We're betting on streaming, but we might be wrong." That clarity enabled faster course corrections.

OpenAI is now changing its grading rubric to reward its agent's capability to help users find the right answer, not providing answers itself.

Leaders can do the same. Handled well, strategic uncertainty creates more value than confident bluffing ever could.

Have you found moments where searching for the right answer earned more respect than knowing it outright?

Could it be that our early conditioning, schools rewarding correctness significantly higher than curiosity, still shapes how we lead today?