How Much AI Risk Is Okay? (That's Risk Tolerance)

In grants work, some expenses needed pre-approval and some didn't. Nobody treated a box of pens like a construction contract. That was risk tolerance in action: deciding in advance how much uncertainty is acceptable for each kind of decision.

AI needs the same conversation. A chatbot drafting internal meeting notes can fail harmlessly. The same model screening job applicants cannot. Your risk tolerance is the line between those two — written down, agreed on by leadership, before anyone is mid-crisis.

Here's a starting point: sort your AI uses into three buckets. Low stakes, where mistakes are cheap and reversible. High stakes, where mistakes hurt people or cost real money. And off-limits, where the risk isn't worth it at any price. Most organizations have never sorted the list. Sort yours this week.

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