When something goes wrong: stop, tell, fix, learn

AI will get something wrong in your work. Not because you used it badly, but because that's what these tools do now and then: a confident wrong figure, a source that doesn't exist, a message that goes to the wrong person. Your organisation isn't asking you never to make a mistake. It's asking you to do four things when one happens.

1. Stop

Stop using it for that task, and don't pass on anything else it made until it's been checked. If something is still happening, such as a feature still sending emails, switch it off if you can do so safely. You don't need anyone's permission to stop something that's going wrong.

2. Tell

Tell the right person within one working day. That's the rule, and it's a deadline, not a target: if you can tell them sooner, do. The right person is whoever your organisation has named for AI problems, often your manager or the person who runs AI work.

If personal or sensitive data is involved (a customer's details, someone's health information, staff records, card numbers), tell whoever looks after that data instead. That's often compliance, or a data protection officer. They're the person the organisation trusts to hold this kind of disclosure, outside your management chain.

Keep the message short. Say four things:

• what happened
• where: the tool, and the piece of work
• what data was involved, if any
• what you've done so far

Describe the work, not a person, even if the person is you. Near misses count too: the wrong figure you caught before it went out is worth telling, because the next one might not be caught.

3. Fix

Fixing it isn't yours alone. The owner of the tool or the way of working puts it right for everyone who uses it: they contact anyone affected, change the setting or the template, and check whatever else the same tool produced. Your part is to help them understand what happened.

4. Learn

Last, a short, blameless note goes on the friction list: what happened, why the work allowed it, and what changes. It records the work, never the person. That's how one person's bad afternoon becomes a check that protects everyone.

Practise on five situations. For each one, choose what you'd do first, then who you'd tell.

Why telling is safe

Most people hesitate to tell. They worry about looking careless, or about what happens next. That worry is common, and it's understandable.

That's why the first of the four promises exists. Telling us is safe. Nothing you disclose about how you use AI is used in performance reviews or in decisions about who leaves. A mistake you disclose is handled through stop, tell, fix, learn, as coaching, not discipline. That holds even if the mistake exposed regulated data or caused harm.

Only wilful or criminal conduct falls outside that protection: knowingly breaking the rules, or breaking the law. An honest slip is never that. And if anyone breaks the promise, that's treated as an incident in its own right.

Aviation learned this decades ago. Pilots who report their own honest mistakes to NASA's Aviation Safety Reporting System are protected from penalties in most cases, as long as the mistake wasn't deliberate or criminal and they report it within ten days. People across aviation file well over a hundred thousand reports a year, and those reports catch problems before they become accidents. Punishing a report doesn't make the next mistake less likely. It makes the next report less likely.

Who you tell, where you work

Now write down your two contacts, so you don't have to look them up on a bad day. If nobody has told you yet, choose “Not told yet”: the question to ask your manager goes on your record. Then practise the message once, so it's easy when it's real.