Are you liable for what AI tells customers about your product?

In most cases the liability sits with you, not the AI vendor. If an AI assistant states a wrong specification, configuration or compliance claim about your product and a customer relies on it, the exposure tends to land on the brand whose product it is — not the model that produced the answer. That is uncomfortable, because you didn’t write the answer and can’t see it being given.

Why does the liability land on the brand?

AI assistants answer by reading public material — your documentation, datasheets, support pages — and reassembling it. To a customer, the answer looks like it came from you. The legal questions that follow (misrepresentation, fitness for purpose, safety information, regulated claims) attach to the product and the party placing it on the market. “The model said it” has not, so far, proven to be a clean defence, and AI vendor terms generally disclaim responsibility for output.

What makes the risk worse

Three things turn this from theoretical into real exposure:

• It’s confident. Wrong answers are stated with the same fluency as correct ones, so customers act on them.
• It’s invisible. The exchange happens in a private chat. You have no log, no record, no chance to correct it.
• It’s unbounded. You can’t predict every question, so you can’t pre-approve every answer.

What reduces it

You can’t guarantee an AI will never be wrong — and you should be wary of anyone selling “absolute accuracy.” What you can build is traceability, evidence, and process: a verification layer that checks what AI says about your products against a current, approved source, flags divergence, and keeps a dated record of what was checked and who signed it off. When something goes wrong, the difference between “we had no idea” and “here is our control, our evidence, and our remediation” is the difference that matters to regulators, insurers and courts.

Where to start

Start with an audit: what do the major AI engines currently say about your highest-risk products, and where does that diverge from your source of truth? That gives you a measured picture of exposure before you decide how much control to build.

This is general information, not legal advice; liability depends on your jurisdiction, sector and the specific facts. Treat it as a prompt to talk to your own advisers.