AI agents are not just chatbots anymore. They answer customers, write code and make small decisions on their own. Gartner expects that by 2028 at least 15% of daily work decisions will be made by agents, without a human. So when an agent makes a mistake, who pays for it?
Courts already gave one answer. In 2024, Air Canada’s chatbot gave a customer wrong information about a discount. The airline argued that the chatbot was responsible for its own actions, but the tribunal did not accept this and Air Canada had to pay.
Most articles I read treat this as a legal problem. One law firm says the developer of the agent is the first place courts look. I don’t fully agree. A register that analysed 12 real incidents shows the company using the agent was liable in 5 cases, the vendor in 4, and both in 3. So there is no single answer, and blaming the vendor will not always work.
For me, this is a management problem before it is a legal one. A Cloud Security Alliance survey found that 82% of organizations discovered AI agents in their systems that they did not know about. How can managers be responsible for something they don’t even know exists?
Gartner also says more than 40% of agent projects will be cancelled by 2027, partly because of weak risk control. I think this is the real lesson. Companies buy agents fast, but nobody decides who “owns” them. You can delegate a task to AI, but you cannot delegate responsibility. Every agent should have a named human behind it before it starts working.
References
- Gartner: over 40% of agentic AI projects will be scrapped by 2027
- American Bar Association: Moffatt v. Air Canada
- The Lyon Firm: Who is legally liable when an AI agent makes a mistake?
- AI Agent Incident Register: Who is liable when an AI agent causes harm?
Prompts:
- “Suggest some blog topics for this assignment about current technology trends and management.”
- “Does the assignment say how many words the post should be?”
- “Can you check my text for obvious grammar mistakes and correct them?”
Strong post, and I like that you frame this as a management problem before a legal one. The incident register detail (company 5, vendor 4, both 3) is a good counter to the idea that you can simply blame the vendor.
Your point about the 82% of organizations finding unknown agents is the most striking. You can’t assign ownership of something you haven’t discovered. I’d add that “a named human behind every agent” works best with an inventory, so shadow agents get found first. Ownership should also include authority to pause the agent and responsibility for reviewing its logs. Otherwise a name on paper means little.