Blog / Liability

Who is responsible when an AI agent decides?

In February 2024 Air Canada argued that its chatbot alone answered for what it told customers. The tribunal refused. The rule that follows is old and simple: a company answers for what the agent it puts into service does, which forces it to know what it has entrusted to that agent.

By Vincent Dorange, , 3 min read

14 Feb 2024

a Canadian tribunal holds Air Canada responsible for what its chatbot told a customer

After his grandmother died, Jake Moffatt looked for a flight to Ontario. The chatbot on Air Canada's website told him he could apply for the reduced bereavement fare after the fact. That was wrong: the airline's own page on that fare excluded requests made once travel was completed, and Air Canada refused the refund. Before the Civil Resolution Tribunal of British Columbia, the airline argued that it could not be liable for what its chatbot had said, presenting it as a separate legal entity responsible for its own actions. On 14 February 2024 the tribunal rejected the argument and found for the customer 1 2.

An agent has no legal personality

The tribunal's reasoning is short: the chatbot is part of Air Canada's website, and the airline is responsible for all the information on its website, whether it comes from a static page or a chatbot 2. The decision comes from a small claims tribunal and is not binding precedent. It still made headlines around the world because it states something many companies would rather forget: an agent is not a person, it has neither assets nor liability of its own, so what it says or does binds whoever put it into service.

French law reaches the same point by another route. Article 1984 of the Civil Code defines a mandate as the act by which one person gives another the power to do something for them and in their name 3. An AI agent is not a mandatary within the meaning of the Code, since it has no legal personality; in practice it nonetheless exercises a power the company has entrusted to it. The company remains the author of what it does, with an added difficulty: a human agent can read the limits of a mandate, whereas a software agent knows only what someone thought to write down for it.

The GDPR already sets a limit

For decisions that affect people, a European rule comes on top. Article 22 of the GDPR gives everyone the right not to be subject to a decision based solely on automated processing that produces legal effects concerning them or similarly significantly affects them, unless the decision is necessary for a contract, authorised by law or based on explicit consent 4. An agent that refuses a refund, rejects a job application or changes a price without human involvement falls within that scope.

Public authorities that have written about agents point the same way. The agentic AI governance framework Singapore presented on 22 January 2026 stresses that humans remain ultimately accountable, and asks organisations to define the checkpoints at which human approval is required 5.

What a company must write down before delegating

If a company answers for everything its agent does, it has to know exactly what it entrusted to it. That is the starting point of the Agentic Commerce Framework®: decisions are delegated to an agent, never the responsibility that goes with them. In practice, four documents should exist before an agent goes live.

  • A written mandate that says what the agent may decide alone, what it must submit to a human and within which figure-based limits, including what it is allowed to promise a customer.
  • The list of non-delegable decisions, those no agent takes, however reliable it appears.
  • A named owner for each agent, along with an overall oversight function the ACF calls the DDAO, Delegated Decision Agent Officer, able to say no and to stop an agent.
  • An audit trail that shows, decision by decision, what the agent did, under which rule and with which approval.

Air Canada lost because it tried to distance itself from its agent after the fact. A company that has written down its delegations before going live does not need that argument: it can show what it had authorised, what it had forbidden and who answered for it.

Sources

Links to the official texts and original publications.

  1. 1Moffatt v. Air Canada, 2024 BCCRT 149, Civil Resolution Tribunal of British Columbia, 14 February 2024.
  2. 2Lifshitz L. R. and Hung R., "BC Tribunal Confirms Companies Remain Liable for Information Provided by AI Chatbot", Business Law Today, American Bar Association, 29 February 2024.
  3. 3French Civil Code, Article 1984 (in French), Légifrance, version in force.
  4. 4Regulation (EU) 2016/679 (GDPR), Article 22, Official Journal of the European Union, 27 April 2016.
  5. 5Singapore Launches New Model AI Governance Framework for Agentic AI, Infocomm Media Development Authority (IMDA), 22 January 2026.

Glossary terms used in this article

AI agent delegationAI agent mandateNon-delegable decisionsAI agent ownerDDAO (Delegated Decision Agent Officer)AI audit trail

Vincent Dorange, creator of the Agentic Commerce Framework®

Also worth reading

95%

of organisations surveyed faced at least one agent-related incident in 12 months

Governance

95% of organisations have already had an AI agent incident, says Gartner

25

organisations interviewed; none lets its agents act with unrestricted autonomy

Governance

AI agents at work: what the OECD learned from 25 organisations

4 levers

transaction caps, budgets, approved counterparties and escalation rules, per Circle

Agentic economy

Agentic economy: agents that spend within delegated permissions

All articles