Liability for AI Agents

Artificial Intelligence (AI) continues to dominate the headlines, with an increasing focus on the use of AI agents. AI Agents are software systems that are capable of performing tasks autonomously on behalf of a human user (or even another AI system).

As AI agents take on more responsibilities, questions arise about who is liable when things go wrong. The 2024 case of Moffatt v Air Canada offers important insights into this issue. Here, we explore the implications concerning liability related to AI agents.

Background of Moffatt v Air Canada

Moffatt v Air Canada involved a passenger named Moffatt, who experienced significant issues during a flight booking process managed partly by an AI system. The AI agent was responsible for handling customer requests, and booking flights. However, errors in the AI’s processing led to a booking mistake, with the AI agent wrongly informing Moffatt that he could retroactively apply for a bereavement discount on his flight tickets. This resulted in Moffatt incurring financial loss and inconvenience.

Moffatt sued Air Canada, claiming the airline was liable for the AI agent’s errors. The court had to decide whether Air Canada could be held responsible for the actions of their AI system, which operated with a degree of autonomy, but was ultimately controlled by the airline.

Court’s Decision and Reasoning

The court ruled in favor of Moffett, holding that Air Canada was liable for their AI agent’s errors in negligently misrepresenting the terms of purchase of flight tickets. The court's reasoning was based on the principle that companies are responsible for the systems they deploy, including AI. The AI agent acted as an extension of the airline’s services, and customers relied on it as part of the company’s operations.

The court emphasized that AI does not have a separate legal identify, and therefore cannot be held accountable independently. Instead, liability falls on the company that controls and benefits from the AI’s actions. This decision aligns with existing legal frameworks where employers are responsible for their employees’ actions, extending the concept to AI agents.

Implications for Businesses Using AI Agents

The ruling of this case sends a clear message to companies using AI agents:

• Businesses must ensure AI systems are reliable and safe.

  Since liability rests with the company, investing in thorough testing and quality control is essential.

• Clear policies and oversight are necessary.

  Businesses should monitor AI performance, and have protocols to swiftly react to address errors.

• Transparency with customers is important.

  Informing users about AI involvement and potential risks can help manage expectations and reduce disputes.

• Legal risk management should include AI considerations.

  Companies may need to update contracts, insurance policies, and compliance measures to cover AI-related issues.

What This Means for Consumers

For consumers, the ruling provides a measure of reassurance that companies remain accountable for AI-driven services. If an AI agent causes harm or loss, customers have a clear path to seek compensation from the company behind the AI.

Consumers should also be aware of the limits of AI and exercise caution when relying on automated systems. Understanding the role of AI in services can help users make informed decisions and report issues promptly.

Some countries have already proactively addressed this issue. For example, the UK government has issued consumer protection guidance, confirming that businesses must ensure compliance with consumer protection laws when using AI agents, just as you would when using human agents.

Conclusion

Nvidia CEO, Jensen Huang, recently stated that he envisaged a future with a ratio of 75,000 humans working alongside 7.5 million AI agents. That amounts to 100 AI agents for every human worker.

The scope for potential liability in a future where we have 75,000 humans instructing 7.5 million AI agents is extensive. However, the question of whether businesses are liable for their AI agents has now been tested in the Canadian courts. This also sits well with the underlying principle that for the creation of protectable intellectual property, there still needs to be human involvement and oversight. As the relationship between AI and humans continues to evolve, hopefully more clarity and guidance will emerge on the responsible use of guardrails in AI systems. Given the scope of development of AI technologies, we may need clearer legal definitions for AI agents and their roles.

Disclaimer: Content is not intended to, and does not constitute, legal advice, and no attorney-client relationship is formed.