Who Watches the Watchers?

Grants took conflicts of interest seriously: nobody with a personal stake in the outcome got to judge it. The reviewer couldn't be the beneficiary. It was a simple recognition that people — even good people — can't objectively grade their own work.

AI oversight has the same blind spot, and most teams haven't named it.

The team that built the system, chose the vendor, or staked its reputation on the launch is now in charge of deciding whether it's working. That's a conflict of interest wearing a lab coat. Of course the review comes back glowing — the reviewers' credibility is on the line.

The grants fix translates directly: separate the builders from the judges. The people assessing the AI system's risks and performance should not be the people whose success depends on it. That might mean a different internal team, an outside reviewer, or simply one person with explicit permission to say "this isn't working" without career consequences.

Independence isn't distrust. It's the thing that makes trust possible.

I explain frameworks; I don't give legal counsel.