The one safety rule

There is one limitation of AI worth understanding properly, because almost every horror story traces back to it. Learn this and you can start using AI today with confidence.

Hallucination, in plain terms

The one word worth knowing is hallucination: when the model states something false with complete confidence. It will invent a statistic, a policy or a fact and present it as certain, because it genuinely has no idea it is wrong.

This is not a bug you can switch off. It is how the thing works. It predicts plausible words, and plausible is not the same as true.

The rule that covers it

From that comes a single rule that handles nearly every situation:

AI drafts, you check.

Never let an unreviewed AI answer reach a customer, a contract or a tax return. Use it to get from a blank page to a solid first version in seconds, which is the genuinely useful part, then you do the checking, which is the part your customers actually pay you for.

What that looks like day to day

• For words (an email, a proposal, some website copy), read it back and fix anything off. Usually quick, because editing is faster than writing from scratch.
• For anything numeric, legal or final, treat the answer as a draft to verify against a real source, never as the last word.
• For anything about your business specifically, remember it is guessing unless you have given it the facts. We cover how to give it those facts in the fuller courses.

That is the whole safety rule. One sentence, and it keeps you out of almost all the trouble. Next up: five real jobs you can hand to an AI this week.