The problem
You run a small agency. You sell a medical supply company a service that checks incoming orders for missing paperwork. It uses an insurance rules list you license from another company. A nurse you contract reviews the hard cases.
One Tuesday, the rules list is a week out of date. Three orders get flagged wrong, and one patient's equipment is delayed. The client calls you. Not the rules company. Not the nurse. You.
So who pays? If you never wrote it down, the answer is usually you.
Why this keeps happening
A business cannot blame its AI and walk away. Air Canada tried in 2024. Its website chatbot gave a customer wrong information about a bereavement fare, and the airline argued the chatbot was responsible for its own actions. A British Columbia tribunal called that "a remarkable submission" and ordered the airline to pay.
The same logic reaches the agency that sold the service. As of July 2026, the law firm Brownstein put it this way: "Responsibility may rest with providers, deployers, integrators, developers, end users or some combination of all of them." An agency that builds and sells AI work can be several of those at once.
Most clients have not sorted this out either. As of July 2026, an Avalara survey of more than 1,500 finance leaders found 23 percent said accountability for a serious AI error would be unclear or would sit with no one.
| Part of the service | Who the client blames | Who should answer, in writing |
|---|---|---|
| The promise and the price | You | You, within the terms you wrote |
| Outside data or rules | You | The supplier, under your contract with them |
| A person reviewing hard cases | You | The reviewer, under your agreement with them |
| The client's own inputs | You | The client, if your terms say so |
How to fix it
Write it down before the next Tuesday. This is general information, not legal advice, so have a lawyer read your final terms. These are the questions to answer.
- State the result in one sentence, and what counts as a failed job.
- Promise only what you control, such as how fast you respond, not how the AI behaves on every input.
- List every part you did not build: whose data, whose rules, and which steps a person handles.
- Get each supplier to back in writing anything you promise the client. If they will not, do not promise it.
- Cap your liability. Brownstein suggests asking: "Is liability capped, and if so, is it reasonable for the amount of risk?"
- Say who covers the cost when something goes wrong, and when you will refund.
- Keep a record of every job, so you can show what happened and when.
- Ask your insurer whether your policy covers mistakes in AI work you sell.
A person checking hard cases only helps if they see the right cases. Why review queues fail, and how to fix them explains how to keep that step working.
What BlueBear does to help
When a client calls, the first question is "what actually happened?" BlueBear keeps a record of each run: when it ran, which tools it used, and what came back. You can answer from the record instead of guessing.
BlueBear workflows can also pause for a person to approve a step before it goes ahead. That lets you put your reviewer exactly where a mistake would be costly.
BlueBear does not take on your legal responsibility to your client. Your contracts still decide that.
What to do next
Answer the eight questions for your busiest service this week. Every blank answer is a risk you are carrying for free. When you change the service later, change management for AI agents covers how to give clients notice.
Want a record of every run your clients' AI makes? Get in touch.
Questions people actually search for
- is an ai agency liable when its ai makes a mistake
Often, yes, at least toward its own client. Tribunals and courts have held businesses responsible for what their AI tells customers. Who ends up paying depends on your contracts with the client and with your suppliers. Have a lawyer review your terms.
- who pays when an ai chatbot gives wrong information
Usually the business that runs it. In Moffatt v. Air Canada in 2024, a British Columbia tribunal rejected the airline's argument that its chatbot was responsible for itself, and ordered the airline to pay the customer.
- what should an ai service contract say about errors
What counts as a failed job, what you promise and what you do not, which parts come from suppliers, how refunds work, and a cap on your liability. Get suppliers to back in writing anything you promise the client.
- how can an ai agency limit its liability
Promise only what you control, cap liability in your contract, get suppliers to back your promises in writing, keep a record of every job, and check that your insurance covers AI work. This is general information, not legal advice.