Skip to main content

SAGANA,BIRIQ&MUGANDA

Menu

Who is responsible when AI causes harm? Courts across the world are already answering that question — and a remarkably consistent thread of reasoning is emerging across very different
cases.

T H E  E M E R G I N G  A N S W E R

“Liability is increasingly being attached not to the algorithm itself — but to the humans, institutions, and companies that design, deploy, supervise, and profit from it.”

When AI causes harm, who should be held responsible? Is it the developer who built the model? The company that deployed it? The employee who relied on it? Or the user who trusted it? Courts and regulators are already confronting these questions — and a consistent thread of reasoning is quietly emerging.

Case 01 — Mata v. Avianca (United States)

U S  F E D E R A L  C O U R T · S D N Y
The Lawyers Who Filed Cases That Did Not Exist
Lawyers submitted court filings containing legal authorities generated by ChatGPT — none of which existed.

The citations were entirely fictitious. The court imposed sanctions on the attorneys after discovering they had failed to verify the material before filing. The court’s reasoning was that lawyers have a professional duty to verify accuracy regardless of the tools they use.

→ Using AI does not eliminate human responsibility. The tool may err; the professional remains
accountable for what they certify.

Case 02 — Air Canada Chatbot (Canada)

C A N A D I A N  C I V I L  T R I B U N A L
Air Canada’s Failed Defense: “Our Chatbot Said It, Not Us”
A customer relied on information from Air Canada’s AI chatbot, which incorrectly stated that bereavement discounts could be claimed retroactively. When the airline refused the refund, it argued it should not be liable for statements made by its own automated system.

The tribunal rejected this defense — businesses cannot escape responsibility merely because misinformation came from an automated system they chose to deploy.

→ If an organization deploys AI to interact with the public, it remains responsible for the consequences of that AI’s actions — regardless of automation.