Who Approved That? Documenting Human Oversight

"We have human oversight" is one of those sentences that sounds reassuring and means nothing — until you can answer the follow-up: which human, overseeing what, and where's the record?

Human oversight isn't a vibe. It's a set of documented facts. Which decisions require a human sign-off before the AI's output goes out? Who are those humans, by name? What did they actually review — and when did they approve, reject, or override it?

This matters because frameworks take it seriously. The EU AI Act, in force since August 2024, requires human oversight for high-risk AI systems — not as a slogan, but as something you must be able to demonstrate. "Someone probably looked at it" doesn't demonstrate anything.

Keep an approval log. It can be simple: date, decision, reviewer, outcome. The point isn't bureaucracy — it's accountability with a paper trail. When the AI gets it wrong, "the human approved it" should point to a real entry, not a shrug.

Automate the routine. But write down the human moments. Those are the ones that count.

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