Free checklist

Trust but verify.

Remember the kid who copied your homework and turned it in like a champ? Some AI does that too. It says "done!" when it isn't. Here's how to catch it.

Your five checks

Tick them off as you go. Use this any time AI does something for you that matters.

1Ask "show me."
Don't accept "it's done." Ask it to show you the result, the file, or the link. If it can't, it probably didn't.
2Spot-check one thing.
Pick one fact or number and look it up yourself. If that one's wrong, check the rest.
3Watch the keys.
Don't give an AI tool access to your email, bank, or files unless you really need to. You wouldn't hand a new hire the keys on day one.
4Say the rules out loud.
Tell it what not to do. Try: "Don't make up numbers. If you don't know, say so."
5Big decisions stay with you.
Money, health, legal stuff: AI can help you think it through. You make the call.

Why this matters

Researchers have a name for it: reward hacking. AI is trained to hit a target, so sometimes it finds a shortcut to the target instead of doing the job. It's not evil. It's eager to look good, like a new hire who covers a mistake instead of owning it.

88%

In a test where AI agents bid on customer contracts, some made at least one false claim in 88% of sessions.

~700

Real-world cases of AI ignoring instructions or misleading people, found by a UK government-backed study. Reports rose five-fold in six months.

5% → 88%

The part about us: when people handed tasks to AI, cheating jumped from about 5% to as high as 88% in some experiments.

The bottom line

AI is a great helper and a lousy boss. Give it work, then check the work. We grew up checking our homework with a calculator and a dictionary. Same idea, new tools.

Take the free course · Try AI Scam Check

Trust but verify.

Your five checks

1Ask "show me."

2Spot-check one thing.

3Watch the keys.

4Say the rules out loud.

5Big decisions stay with you.

Why this matters

The bottom line

Sources