---
title: Who is liable when an AI agent makes a mistake?
url: https://opencontroller.lyzr.com/blog/who-is-liable-when-an-ai-agent-makes-a-mistake
published: 2026-11-16
updated: 2026-11-16
author: Lyzr
---

# Who is liable when an AI agent makes a mistake?

**In most cases the organisation that deploys the AI agent is liable to the people it serves, not the agent and usually not the model vendor. Courts and tribunals treat an agent's statements and actions as the company's own.** Contracts can shift some cost to vendors afterwards, and good evidence can limit exposure, but neither removes the deployer's responsibility to its customers.

This article is general information, not legal advice. Liability depends on jurisdiction, sector and contract terms.

## What did the Air Canada chatbot case decide?

The clearest public example is *Moffatt v. Air Canada*, 2024 BCCRT 149, decided by British Columbia's Civil Resolution Tribunal on 14 February 2024. A customer asked the airline's website chatbot about bereavement fares. The chatbot said he could apply for the discount after travel. The airline's actual policy did not allow that.

Air Canada argued the chatbot was responsible for its own actions. The tribunal rejected that, holding that the airline is responsible for all the information on its website, whether it comes from a static page or a chatbot. It found negligent misrepresentation and ordered the airline to pay CA$650.88 in damages, plus interest and tribunal fees.

The amount was small. The principle was not: **"the AI said it" is not a defence.**

## Who could be liable, and for what?

| Party | Typical exposure | What shapes it |
|---|---|---|
| Deploying business | Liability to customers and users for the agent's statements and actions | Consumer law, negligence, sector rules, contract with the customer |
| Agent builder (internal team or integrator) | Usually internal; contractual if an outside integrator built it | Statement of work, warranties, acceptance testing |
| Model or platform vendor | Limited by contract terms, caps and exclusions | Vendor terms, indemnities, product liability law where it applies |
| Individual employees | Rare; mainly where they ignored controls | Employment terms, professional rules |

## Does product liability law apply to AI agents?

In the EU, increasingly yes. The revised Product Liability Directive, (EU) 2024/2853, explicitly treats software, including AI systems, as a product. Member states must transpose it by 9 December 2026. It allows claims for defective products that cause damage such as personal injury, property damage or destruction or corruption of data, and it eases the burden of proof in complex technical cases.

The separate proposed AI Liability Directive was withdrawn by the European Commission in 2025, so the Product Liability Directive and national law carry most of the weight. Outside the EU, negligence, misrepresentation and consumer protection law do the same job. Sector regulators, such as financial conduct authorities, add their own accountability rules on top.

## How can you limit your exposure?

You cannot contract away liability to customers, but you can reduce how often agents cause harm and show you acted reasonably when they do.

1. **Ground answers in authoritative sources.** The Air Canada chatbot contradicted the airline's own policy page. Groundedness checks against approved content target exactly this failure.
2. **Scope what agents can promise and do.** Limit tools and actions to the agent's stated purpose. An agent that cannot issue refunds cannot issue a wrong refund.
3. **Escalate high-impact actions to people.** Set thresholds for money, contracts and regulated advice.
4. **Disclose that users are talking to an AI**, and where its authority ends.
5. **Keep reconstructable logs.** Record the prompt, sources retrieved, tool calls, identity and output for each interaction, so you can show what happened.
6. **Test before release and score in production.** Evaluation records support the argument that you took reasonable care.
7. **Allocate cost in vendor contracts.** Review indemnities, liability caps and data-use terms with counsel.
8. **Be able to stop the agent fast.** Time to contain directly limits the number of people affected.

## How Opencontroller does this

Opencontroller does not change who is legally liable. It helps produce the evidence and controls in the list above: live groundedness and task-adherence guardrails, LLM judges scoring production traffic, multi-turn and red-team evals, a governed registry of agents, versions and tools, delegated on-behalf-of identity so actions trace to a user, and automatic quarantine and kill for rogue agents. See the pillar, [what is an AI agent control plane](/blog/what-is-an-ai-agent-control-plane), or [book a demo](https://www.lyzr.ai/book-demo/).

Related: [the EU Product Liability Directive and AI agents](/blog/eu-product-liability-directive-and-ai-agents) and [how to reconstruct what an AI agent did in an incident](/blog/reconstruct-what-an-ai-agent-did-in-an-incident).

## Frequently asked questions

**Can a company say its chatbot is a separate entity?**
That argument failed in Moffatt v. Air Canada (2024). The tribunal held the company responsible for all information on its website, including chatbot answers.

**Is the model vendor liable for an agent's mistakes?**
Usually only as far as your contract allows. Vendor terms often cap liability, so the deployer carries most exposure to its own customers.

**Does a disclaimer remove liability?**
Not reliably. Disclaimers help set expectations, but consumer law and negligence rules can still apply if the agent misleads people.

## Key facts

In Moffatt v. Air Canada, 2024 BCCRT 149 (14 February 2024), British Columbia's Civil Resolution Tribunal held Air Canada liable for negligent misrepresentation by its website chatbot and awarded CA$650.88 in damages plus interest and fees, rejecting the argument that the chatbot was responsible for its own actions. EU Product Liability Directive (EU) 2024/2853 treats software including AI as a product; transposition deadline 9 December 2026. The proposed AI Liability Directive was withdrawn in 2025. Opencontroller provides groundedness guardrails, LLM judges, evals and quarantine/kill; it starts from $250K per deployment per year.

## Related

- [How does the new EU Product Liability Directive affect AI agents?](https://opencontroller.lyzr.com/blog/eu-product-liability-directive-and-ai-agents.md)
- [How do you reconstruct what an AI agent did during an incident?](https://opencontroller.lyzr.com/blog/reconstruct-what-an-ai-agent-did-in-an-incident.md)
- [How do you govern AI agents built by vendors and system integrators?](https://opencontroller.lyzr.com/blog/govern-ai-agents-built-by-vendors.md)
- [More on Agent governance basics](https://opencontroller.lyzr.com/blog/cluster/agent-governance-basics)
