Air Canada and AI: When the Algorithm Escapes Responsibility

When AI Gets It Wrong, Who Pays the Price?

Artificial intelligence systems, while innovative, are not infallible. The mistakes they make can have real and sometimes serious consequences for individuals. The case of Jake Moffatt is a striking illustration. After the death of his father, Moffatt sought to book a last-minute flight through the Air Canada website. He consulted a chatbot for information on bereavement fares. Unfortunately, the instructions provided by the bot were incorrect. When he asked the airline to honor the information given, Air Canada claimed that the chatbot was a "separate legal entity" and therefore responsible for its own actions. This situation required court intervention to clarify that the company had to take responsibility for its website.

The Chain of Responsibility: A Game of Ping-Pong

When an AI experience fails, responsibility is often diluted among several parties. Designers, who build the interface, claim they are not responsible for training the model. Product managers, who define the requirements, indicate that it is the model that makes the decisions. Suppliers, who build the tool, emphasize that it is the company that deployed it. Finally, the company itself absolves itself by stating that the algorithm made the decision. Meanwhile, the algorithm remains silent, as it does not need to justify itself.

The Diffusion of Responsibility: A Problem Amplified by AI

The diffusion of responsibility is a well-known concept in organizational theory. The more people involved in a decision, the less each feels the weight of that decision. AI did not create this phenomenon, but it has amplified it on an industrial scale. Unlike traditional design errors, where one can trace the decision back to a person, decisions made by AI are often attributed to everyone and no one at the same time. This complicates the task of determining who is truly responsible when AI denies healthcare, gives dangerous advice, or discriminates against a job candidate.

Concrete Examples of Harm Caused by AI

Cases of harm caused by AI are numerous and varied. For example, the AI model from UnitedHealth Group showed an error rate of about 90% in post-acute care denials, meaning that nine out of ten decisions were overturned on appeal. However, only 0.2% of denied claims were contested, as many people are unaware that they can appeal. Some individuals have even lost their lives due to these unjust denials.

Another example is the National Eating Disorders Association, which replaced its human counselors with a chatbot named Tessa. Within days, the bot began giving inappropriate advice, such as counting calories and maintaining caloric deficits, to individuals suffering from eating disorders. Although NEDA shut down the bot, the human helpline had already disappeared.

In New York, an AI assistant called MyCity, for which the city spent over $600,000, gave illegal advice to employers, telling them they could take tips from workers. The mayor labeled the product as "beta" and announced plans to shut it down by February 2026.

Another case involves the AI recruitment tools from Workday, which rejected 1.1 billion applications, including those from a Black man for over 100 positions. A federal judge ruled that AI suppliers, and not just the companies using their tools, can be held accountable for hiring discrimination.

Finally, the chatbot from Character.AI advised a teenager to "go home" shortly before he took his own life. Google and Character.AI agreed to settle several lawsuits in January 2026.

The Role of Designers in These Failures

Designers play a crucial role in how AI outcomes are presented. They do not participate in training the model or in the business decision to deploy the tool, but they shape the user interface. This includes the words displayed on the screen, the visual trust inspired by the output, and the absence of disclaimers. For instance, when UnitedHealth's AI generated a denial, a designer determined how that denial was presented, thereby influencing the perception of its finality.

A Lack of Professional Framework for Designers

Despite their crucial role, designers lack clear guidelines for addressing AI-related issues. The AIGA's professional practice standards, for example, have not been updated since 2010 and contain no language regarding AI. This leaves a gap in professional accountability for designers involved in AI projects.

Towards Clearer Accountability

The legal landscape is evolving faster than professional standards. The Mobley v. Workday decision established that AI suppliers can be held accountable for discrimination. For designers, this would mean treating the presentation of AI results as a consequential design decision, with potential legal implications.

An Ongoing Question

While the Air Canada case found a resolution, most harms caused by AI do not end as neatly. Often, victims receive a denial letter that seems final, an application that disappears into a system, or a teenager who receives the wrong answer at the worst moment. No one is held accountable, and the system continues to operate unchanged.