Skip to main content
Guides & How-To

Who Is Liable When AI Makes a Mistake in a Business?

As a starting point, a company bears the responsibility itself for what it puts into operation, even when the work is carried out by an AI. An AI solution is a tool in your processes, on a par with an employee or a system, and towards customers, authorities and partners it is you who answer for the result. So the answer isn't "the AI takes responsibility" but "build the process so that a person stands behind the decisions that matter". That is what this article is about.

Author: Agent360 editorial team
Published: 9 September 2026
Last updated: 9 September 2026
Reading time: 7 min read

Important first: This is not legal advice. Specific liability always depends on the specific situation, your agreements and the rules that apply, and should be assessed by a lawyer when something is serious or unclear. What follows is an honest, practical walk-through of how to think liability through, so you go into AI with your eyes open.

The basic principle: you are liable for your own processes

The most widespread misunderstanding is that responsibility "moves over" to the technology or to the supplier when you automate something. As a starting point, it doesn't.

When your company sends a quote, replies to a customer, posts an invoice or makes a decision, it is the company that acts: whether it is an employee, a spreadsheet or an AI that carried out the task. An AI that makes a mistake is in that sense not so different from a new employee who makes a mistake: it is still the company's name on the result, and it is the company the customer holds responsible.

That also means that responsibility doesn't disappear because you didn't write the specific sentence yourself or carry out the specific action. If you have put a process into operation, you are also responsible for it working properly, and for stepping in if it doesn't.

The good news is that this is exactly the same logic you already use for everything else in the business. You have routines for who approves what, when something should be double-checked, and when a decision is big enough to go up the management chain. AI doesn't change the principle: it just makes the work faster, and therefore places sharper demands on having thought the control through in advance.

Human control over the decisions that matter

Because responsibility stays with you, the single most important move is to decide where a person has to be in the loop. It isn't about having a person approve everything (then you lose the whole gain) but about being deliberate about which decisions are critical enough to require approval.

A useful way to divide it up:

Low risk: AI can run on its own

Internal drafts, suggestions, sorting, summaries, first versions. Here the consequence of a mistake is small and easy to correct, and speed is the whole point. A person sees the result, but doesn't need to approve each individual step.

Medium risk: a person approves before sending

Customer-facing communication, quotes, texts that leave the building. The AI produces the draft, a person reads it through and presses send. It takes a few seconds and removes the vast majority of embarrassing or expensive mistakes.

High risk: a person makes the decision, AI supports

Agreements with financial commitments, matters involving sensitive personal data, legal or security decisions. Here the AI should supply the basis, overview and drafts, but the decision rests with a person who can be held accountable for it.

The point isn't to slow the AI down, but to place the human control where a mistake actually hurts. It is both the responsible approach and the one that gives you the peace of mind to use AI widely for everything else.

Supplier agreements and data processing agreements

When you use an external AI supplier, another layer arises: the relationship between you and the supplier. Here two things are worth having a handle on.

The supplier agreement itself describes what the supplier stands behind, what they disclaim, and how responsibility is divided between you. Many standard agreements for software services place a large part of the responsibility for the use with the customer, that is, you. So read what it actually says before you assume the supplier will "cover" mistakes. If something is unclear or important, it is an obvious place to have a lawyer look it over.

The data processing agreement is relevant if the AI solution processes personal data on your behalf, for example customer data. Under the data protection rules, it must be clear who processes which data, for what purpose, and how it is protected. It is an area where it rarely pays to guess; get it in place in writing, and involve advice if you are in doubt about whether a particular agreement is sufficient.

A sound rule of thumb: assume that responsibility for how you use the solution stays with you, unless an agreement expressly says otherwise. That keeps you on the cautious side.

Insurance: ask your own insurer

A natural question is whether your existing business insurance covers damage arising from an AI error. The honest answer is: only your insurance company knows.

Insurance cover depends entirely on the specific policy, your industry and the specific incident. We can't (and no supplier should) promise you that "it's covered". What you can do is contact your insurance company or broker and ask the question directly: How does our current cover relate to errors that arise when we use AI in our processes? It is a cheap conversation to have in advance and an expensive realisation to get afterwards.

So treat insurance as an active clarification, not an assumption.

Documentation and traceability

Wherever liability ends up, you stand far stronger if you can show what happened and that you acted sensibly. Traceability is both good operations and a safeguard if something goes wrong.

In practice that means being able to answer:

  • What was decided, and by whom? Can you see whether a person approved, or whether something went through automatically?
  • What was the decision based on? What information did the AI have when it produced its draft?
  • When did it happen? A simple timeline makes a big difference if a case has to be reconstructed.

It doesn't have to be heavy. Often it is enough to log the critical steps (who approved what, when) rather than trying to save everything. The purpose is that you can document that the process was sound, and quickly find out where a mistake arose, if one does.

When should you ask a lawyer?

You can think a lot through yourself. But there are situations where professional legal advice is worth the money, and where it is downright irresponsible to settle for an article like this one:

  • When the AI is part of decisions with significant financial consequences or binding agreements.
  • When sensitive personal data is processed and you are unsure of the rules.
  • When you are about to sign a supplier or data processing agreement you don't fully understand.
  • When a mistake has already happened, and a real dispute or claim is on its way.
  • When the AI touches a regulated area in your industry.

Involving a lawyer at the right moments isn't a sign that AI is dangerous, it is the same common sense you apply to every other significant decision in the business.

How we do it at Agent360

At Agent360 we build AI solutions with human-in-the-loop as a basic premise, not an afterthought. We help you decide where the AI can safely run on its own, and where a person must approve before anything leaves the building, so you get the speed without letting go of control over what matters.

We advise honestly: we don't promise you that liability or insurance is "solved", because that depends on your situation. But we build your processes so that they are well thought through, traceable and sound, and so that you know exactly where the human sits in the decisions.

Book a no-obligation AI consultation with Agent360. We look at your processes, talk openly about responsibility and control, and find the places where AI gives the most value with the least risk.

This article is general information, not legal advice. Specific liability depends on the agreement and the situation; seek professional advice if in doubt.

What takes up your time?

Write down the task that takes up the most time. We reply whether it is worth building, and we say so if it is not.