Your Approve button might like the cookie banners we all ignore.
Every AI policy says “human oversight required.” Most launch checklists tick it with one box. Most products build it as one Approve button.
That button quietly assumes the answers to 6 questions nobody asked.
- Is the task appropriate?
- Does the screen show what’s needed to say no?
- When does attention run out?
- How would a failure surface?
- Where does the blame land?
- What’s quietly getting worse?
Under those six questions sit 51 known ways human review fails: automation bias, alarm fatigue, decision fatigue, the moral crumple zone, normalization of deviance, and 46 more.
In most products, none of them have been specified, tested or measured. A ticked box says a button exists. It doesn’t say it works.
There are 51 ways it might not be.