The Failed Fix of Disclosure
Last time I argued that employees are turning into readers of AI output. The common fix: require people to label what's AI and what's their own. And the immediate objection: what if they lie? If your system depends on honest disclosure that you then have to verify, it will fail. AI detectors don't work, the harder you police the deeper people hide it, and you torch trust along the way.
Making the Lie Pointless
The real fix isn't stopping the lie. It's making the lie pointless. Move evaluation from the deliverable to the defense of it. Why did you decide this? What happens when this recommendation actually lands? AI can't sit in that conversation for them.
Two Flavors of Cheating
Once you do that, you'll notice cheating comes in two very different flavors. One is lying about authorship: AI wrote it, they claim they did, but they genuinely understand it, can explain it, and will own the outcome. That's a vanity problem. It barely touches results. The other is the dangerous one: passing off AI output they don't understand as their own judgment. And that one gets caught by real consequences downstream, not by a detector. The accountability sits squarely on them.
Don't Punish the Honest Majority
So don't put the whole company under surveillance to catch the few who lie. That punishes the honest majority — and pushes even more people into hiding. Don't try to stop people from lying. Build a system where lying doesn't pay.
Takeaways
The harder you police the deeper people hide it, and you torch trust along the way.
AI can't sit in that conversation for them.
The accountability sits squarely on them.