Who Is Responsible When AI Makes a Mistake?

“Is AI liable when it makes a mistake?” is the wrong question. AI has no bank account, no malpractice insurance, and no incentive that changes when it gets something wrong. It cannot be sued, fired, or held in contempt of court. The question that actually matters is: who deployed this system, who reviewed what it produced, and who put it in front of the person who got hurt by the error. Answer that, and you have found the responsible party. In nearly every real-world scenario a builder will face, responsibility sits with whoever shipped the system, not with the model vendor underneath it.

That answer holds whether you are a solo freelancer bolting a chatbot onto a client's site or a founder running an AI feature at scale. Courts and contracts generally treat AI output the way they would treat output from any other tool or employee: the operator who chose to deploy it, and who profited from deploying it, owns what happens next. For a broader map of where AI failures actually show up in real products, see this overview of common AI risks builders run into.

The Freelancer Who Ships an AI Feature for a Client

When a freelancer wires an AI feature into a client's product, a support bot that quotes the wrong refund policy, an AI-drafted contract clause that turns out to be unenforceable, an image generator that spits out something the client can't legally use, the client does not chase the model vendor. They chase whoever they paid. In practice, that is you. Freelance contracts rarely spell out AI-specific liability, so when something breaks, the default assumption is that the person who built and delivered the thing is on the hook for it.

Put AI use in the statement of work, in writing, before you ship. Name which parts of the deliverable are AI-generated or AI-assisted, what review you did, and what the client is responsible for checking before launch. This is also where disclosure works in your favor, not just as an ethics question but as a liability one. Being upfront about where and how you used AI, and getting the client to sign off on it, moves part of the review burden onto them and gives you a paper trail if something goes wrong later.

The Agency Deploying AI for Multiple Clients

An agency running AI across ten client accounts carries ten times the exposure of a freelancer, without the built-in slack that comes from being a single point of contact. If one client's AI-generated ad copy makes a false claim, or a chatbot mishandles a support ticket that turns into a complaint, the agency's name is on the invoice and usually on the contract too. Agencies also sit closer to regulatory scrutiny than solo freelancers do, especially as rules like the EU's high-risk AI provisions start applying to systems used in hiring, credit, or other sensitive categories.

Keep an audit trail per client: which AI tool was used, what prompt or configuration drove the output, and who reviewed it before it went live. If you are deploying anything that touches hiring, lending, health, or other high-stakes categories, price out professional liability insurance that explicitly covers AI-assisted work. Most general errors-and-omissions policies were not written with this in mind, and some insurers now exclude AI-related claims unless you ask for a rider.

The Solo Founder Running an AI SaaS

If you are a founder shipping an AI product directly to users, you are the last stop before the mistake reaches someone. Your terms of service can cap your liability and disclaim warranties, but they cannot make the responsibility vanish, especially if the harm involves unauthorized advice, a financial calculation error, or an infringing output your product generated. Model vendors write their terms specifically to push this liability downstream to you. The API is a tool, and you decided what to build with it and who to sell it to. That extends to intellectual property questions too, including who actually owns what the model outputs and what you are allowed to do with it commercially.

For anything with real consequences, money movement, medical or legal-adjacent advice, irreversible account actions, keep a human in the loop before the action executes, not just a human who reviews it after the fact. That single design choice is the difference between catching a bad output before it ships and explaining to a user why your AI told them something false after the fact.

The Employee Using AI Tools at a Company

If you are an employee using AI tools on the job, in most cases your employer carries the external legal and financial consequences, not you personally. That is generally how employment law treats mistakes made within the scope of a job. But that does not mean you are insulated. If you used a tool the company had not approved, skipped a review step that was actually required, or pasted confidential client data into a public chatbot to get a faster answer, you can face internal consequences: a bad review, a policy violation on your record, or termination, even while the company absorbs the external liability.

Know your company's actual AI policy, not the version you assume exists. If there is not one, do not treat that as permission. Ask what is approved for client data and high-stakes outputs before you rely on it, and keep a light paper trail, an email or a message, showing you flagged AI-assisted work where it mattered.

What This Doesn't Cover

This is a practical map, not legal advice, and it cannot be. Liability rules differ by jurisdiction, by industry, and by contract, and regulated fields like healthcare, finance, employment, and anything touching minors often have specific statutes that override the general pattern described here. Criminal liability, for fraud or discrimination facilitated by an AI system, follows its own separate and much stricter logic. If an AI mistake has already caused real damage, or you are structuring a business around AI in a way that carries real exposure, talk to a lawyer who knows your jurisdiction and your industry. This piece is meant to help you think about where the risk sits before you need one.

FAQ

Can I sue an AI company when its model gives me bad advice?

In most cases, no, or at least not successfully on its own. Model vendors' terms of service typically disclaim liability for output and tell users not to rely on it for high-stakes decisions without review. The party you would realistically have a claim against is whoever built the product you were using and put that advice in front of you without adequate warnings or review.

Is a company liable for what its AI chatbot tells customers?

Generally yes. When a company deploys a chatbot on its own site or product, its statements have tended to be treated the same as statements from any other company representative or system. The company chose to deploy it, controls its configuration, and profits from the interaction, so it typically owns the consequences of what the bot says.

Do freelancers need insurance to use AI in client work?

It depends on the stakes of the work. For low-risk deliverables, a clear contract and honest disclosure may be enough. For anything touching client finances, health information, hiring decisions, or legal content, professional liability insurance that covers AI-assisted work is worth the cost, because a single client dispute can outweigh years of premiums.

What happens if an employee uses unauthorized AI tools at work?

The company usually still carries the external liability if something goes wrong, since the employee was acting within their job. Internally, though, the employee can face real consequences: disciplinary action, a formal write-up, or termination, particularly if the unauthorized tool exposed confidential data or the company has an explicit policy against it.

Does disclosing AI use to clients reduce legal risk?

Yes, in two ways. It shifts some of the review responsibility onto the client once they have agreed to how AI was used, and it creates a paper trail showing you were not concealing anything if a dispute happens later. Silence protects no one. If a client finds out after the fact that AI was involved and was not told, that discovery itself becomes part of the dispute.