For the past two years, AI customer service failures have mostly circulated as embarrassing one-off stories: a swearing chatbot, a fabricated discount, a bot that invented company policy. In 2026, the pattern moved from anecdote to data. Consumers rank AI applications for customer service among the worst for convenience, time savings, and usefulness, and a failure rate almost four times higher than for AI use in general now shows up in survey research rather than just viral screenshots.

The numbers are specific and uncomfortable for anyone running a support operation. The Qualtrics 2026 CX Trends Report found that nearly one in five consumers who used AI for customer service got no benefit from it, a failure rate the report frames as several times higher than AI use in general. A separate survey was worse: Glance's 2026 CX Trends Report, a survey of more than 600 US consumers, found that 75 percent had received a fast AI-driven response that still left them frustrated, that about a third said AI support made things harder, and that nearly 90 percent reported reduced loyalty when human support was removed. Both of those reports come from vendors with a stake in the CX space, which is worth keeping in mind, but the direction of the signal lines up with what courts have already been ruling on for two years.

That legal exposure is not hypothetical. The foundational case remains Moffatt v. Air Canada, where the British Columbia Civil Resolution Tribunal ruled Air Canada liable for a chatbot's fabricated bereavement fare policy, after Jake Moffatt consulted Air Canada's chatbot following his grandmother's death and was told he could buy a full-fare ticket and claim a retroactive bereavement discount within 90 days. Air Canada's defense effectively argued the bot was on its own; the tribunal member wrote that "In effect, Air Canada suggests the chatbot is a separate legal entity that is responsible for its own actions," calling it "a remarkable submission". That was not a one-time fluke. Reporting on the pattern since notes a similar case at Cursor in April 2025, where a support bot fabricated a device-limit policy that did not exist, echoing Air Canada almost exactly a year later.

The industry's own leaders concede the incentive structure is the actual problem, not the underlying model. Decagon CEO Jesse Zhang, whose chatbot company tripled its valuation on the back of enterprise deals signed in 2025, put it plainly: "We have not come across a single customer with the intention of deflection," and companies aiming to deflect customers "will lose money in the long run". In other words, the technology isn't the villain in these stories, the scoreboard is. A bot tuned to close tickets fast and cheap will generate exactly the plausible-sounding, unverified answers that cost Air Canada real money in court, while a bot tuned for genuine resolution can avoid that trap because it's allowed to say "I don't know, let me check."

For any business running or considering a customer-facing AI tool, this week's action is a scorecard audit, not a shutdown. Pull whatever KPI currently grades your chatbot, whether that's containment rate, average handle time, or cost-per-contact, and check whether "resolved the customer's actual problem" is measured anywhere at all. If it isn't, you are managing for the metric that got Air Canada sued, not the one that keeps customers loyal. Any statement your bot makes about money, refunds, or entitlements should come from a hard-coded, human-approved template rather than a generated response, and every promise the bot makes, not just its errors, should be logged and reviewable. That single change costs a fraction of what a fabricated policy costs in court, in customer trust, or in the viral screenshot that shows up on your CEO's desk.

Sources