"Who Is Responsible for AI Mistakes?" What Happens When It Gets It Wrong

Short answer: you are, if you’re the one using the answer. Longer answer: it depends on who put the AI there in the first place — and that’s worth understanding, because it changes what you should check before you act on anything it tells you.

An AI can’t be held responsible for anything, in the way that word usually works. You can’t send it a bill, take it to court, or ask it to apologise and mean it. It has no bank account, no job to lose, no reputation to protect. So when something goes wrong, the responsibility has to land on a person or a company — and working out which one matters more than it sounds.

Why “blame the AI” doesn’t work

Responsibility, in the ordinary sense, needs someone who could have done otherwise and who answers for the outcome. An AI chatbot doesn’t meet either test. It didn’t choose to get something wrong the way a careless tradesman chooses to cut a corner. It produced text that looked plausible, based on patterns in its training, with no understanding of whether it was true.

That’s why every AI company’s terms of service say, in effect, the same thing: check the output before you rely on it. It’s usually written in small print you’ve scrolled past, but it’s the honest position. The company built a tool that’s often right and sometimes confidently wrong, told you so in the terms, and left the checking to you.

Two different situations

It helps to split this into two cases, because the answer is different for each.

1. You’re using AI yourself — asking a chatbot to explain a letter, draft an email, or answer a question out of curiosity. Here, you’re the one deciding what to do with the answer. If you act on something wrong — pay a bill you didn’t owe, send an email with an error in it, believe a fact that turns out to be false — that’s on you, in the same way it would be if you’d misread a leaflet or misheard advice from a stranger. The AI made a mistake; you’re the one who acted on it without checking.

This isn’t meant to sound harsh. It’s the same rule that already applies to anything you read online. The difference is that AI sounds far more confident than a stranger’s opinion, which is exactly why the checking step matters more, not less.

2. A company puts AI in front of you — a bank’s chatbot, an airline’s booking assistant, a council’s automated helpline. Here, you’re not choosing to experiment with a new tool; you’re being served by one, often without much choice in the matter. If that chatbot gives you wrong information and you act on it in good faith — it confirms a refund that never arrives, say, or gives you an answer that turns out to be against the company’s own policy — the company is generally on the hook, not you. They chose to put that system in front of customers, and they’re responsible for how it behaves, the same as if a human employee had told you the wrong thing.

The practical difference: keep a record. A screenshot of the chat, or the date and roughly what it said, turns “the chatbot told me something different” into something a human employee can actually act on when you follow up.

What this means day to day

None of this is a reason to avoid AI — it’s a reason to treat its answers the way you’d treat advice from a knowledgeable but occasionally unreliable acquaintance: useful, worth listening to, but not the final word on anything that matters.

A few habits make this easy:

• For anything with a consequence — money, health, a legal letter, a decision you can’t easily undo — use the AI’s answer as a starting point, then check it against something official: the actual bill, the actual doctor, the actual terms and conditions.
• Ask it to show its working. “Where does that come from?” or “how sure are you?” often exposes a guess dressed up as a fact. Here’s more on spotting when it’s wrong.
• Treat a company’s chatbot like a company employee, not like magic. If it gives you something in writing — a reference number, a promised refund, a date — keep it, the same way you’d keep an email.

When to bring in a person

Some situations are worth a human from the start, not a chatbot with a human as backup:

• Money above what you’d want to lose without a second opinion — get a person at the bank, or an independent adviser, to confirm anything an AI suggested before you act on it.
• Legal matters — a chatbot can help you understand a letter, but it can’t represent you or guarantee it has the current law right. For anything with real stakes, that’s a solicitor or a free advice service, not a chatbot.
• Health decisions — AI can help you prepare questions for a doctor. It should never be the thing that decides whether you see one.

The complaint route still works

If a company’s AI gave you wrong information and it cost you something, complain the way you always would: contact the company, explain what happened and when, and ask them to put it right. The fact that a machine was involved rather than a person on the phone doesn’t remove the company’s responsibility for its own service — if anything, most companies are quite aware that a badly behaving chatbot is a problem they need to fix, and complaints are how they find out.

If you don’t get anywhere, the ombudsman or regulator for that industry (banking, energy, telecoms — whichever applies) is the next step, exactly as it would be for any other complaint. What AI actually is is worth reading first if any of this still feels like a black box — knowing roughly how these systems work makes it much easier to judge when to trust one and when to double-check.

The short version

• An AI can’t be held responsible for anything — it has no job, no money, nothing to lose. Responsibility always lands on a person or a company.
• If you’re using AI yourself out of curiosity, checking the answer before you act is down to you.
• If a company puts AI in front of you as part of its service, the company is responsible for what it tells you — keep a record and complain the normal way if it gets something wrong.
• For money, legal or health decisions, use AI to prepare, not to decide. Bring in a real person for anything that matters.
• A confident-sounding answer isn’t the same as a correct one. That’s true of AI, and it was true of confident-sounding people long before AI existed.