What Legal and Liability Exposure Do You Take On When Your AI Agent Gives a Customer Wrong Information?

Quick answer

Courts and regulators have generally treated statements made by a company's AI agent the same as statements made by any other company representative or channel — the company is on the hook, not the AI vendor or the model provider, and "the AI made a mistake" has not been accepted as a defense in the cases that have gone this direction publicly. The practical mitigation isn't legal boilerplate alone; it's constraining what the agent is allowed to assert (especially on price, policy, and contractual terms), logging every interaction for auditability, and having a documented correction process for when it's wrong.

Precedent already exists, and it isn't in the company's favor

A well-known case involved an airline's support chatbot inventing a bereavement-fare policy that didn't exist; the airline argued the chatbot was a separate legal entity responsible for its own actions, and lost — the tribunal held the company responsible for what its chatbot told the customer, full stop. Treat this as the baseline assumption: whatever your agent tells a customer, your company owns it, regardless of whether the answer came from a hardcoded response or a model's generated output.

The risk is highest exactly where agents are most useful: pricing and policy

Agents are valuable precisely because they can answer specific, contextual questions instantly — but pricing, refund policy, contract terms, and legal/compliance questions are the categories where a wrong answer creates the most direct financial and legal exposure. This is a strong argument for hard-gating those categories: retrieve pricing and policy answers from a verified source of truth rather than letting the model generate or reason its way to a number, and route genuinely ambiguous cases to a human rather than letting the agent guess.

Explainability isn't just a compliance nice-to-have here

If a wrong answer does get disputed, being able to reconstruct exactly what the agent knew and why it said what it said is the difference between a quick resolution and an expensive one. This is the direct business case for making an AI agent's reasoning explainable for auditors — not an abstract ethics requirement, but the evidence trail you'll need if a customer or regulator challenges a specific interaction.

Logging and audit trails are your primary defense, not your terms of service

A disclaimer buried in your terms of service ("AI-generated responses may be inaccurate") provides some protection but is weaker than most companies assume, especially against consumer protection claims. What actually protects you in a dispute is a complete, timestamped log of what the agent said, what context it had access to, and what correction process followed — the same operational discipline covered in handling AI agent errors and hallucinations gracefully. If you can't reconstruct an interaction after the fact, you can't defend it.

Build a correction and honoring process before you need one

When an agent does give wrong information a customer reasonably relied on — a quoted price, a stated policy — decide in advance whether you'll honor it, and under what conditions. Companies that have a clear, pre-decided policy (e.g., honor reasonable pricing errors below a threshold, correct and apologize for the rest) handle these incidents faster and with less reputational damage than companies improvising a response in real time under public pressure.

This is a legal and compliance review item, not just an engineering one

Wrong-answer liability should be an explicit line item in the legal and compliance sign-off process before launching a customer-facing agent — not something legal discovers after a customer complaint. Bring legal in during design, specifically to review which categories of question the agent is allowed to answer freely versus which must be hard-gated to verified sources.

---

FAQ

Can a company avoid liability by stating the responses are "AI-generated" in a disclaimer? A disclaimer helps but isn't a complete shield — regulators and courts have generally still held companies responsible for what their AI agent tells customers, particularly when the customer had no reasonable way to know the information was wrong.

Should an AI agent ever quote binding prices or policy terms directly? Prefer retrieval from a verified, single source of truth over model-generated numbers; if the agent can't retrieve a confident, sourced answer, it should say so rather than estimate.

Who is liable if the wrong answer came from a third-party model provider's error, not a bug in your code? Your company, in essentially every documented case so far — the model provider's terms of service typically disclaim responsibility for output accuracy, which is why gating high-stakes answers matters more than trusting the model.

Does this liability risk apply to internal-facing agents too? Less directly — internal agents used by employees carry lower external liability but still create risk if employees act on wrong information in customer-facing decisions, which is one reason internal and external agents need different review standards.

Accelate designs customer-facing agents to hard-gate pricing, policy, and legal answers to verified sources, with full audit logging built in from day one.