A support reply can become a product crisis when customers treat it as official policy. That is what happened after Cursor users saw sessions end when moving between machines and received an authoritative answer from an AI support bot named Sam.
The problem was not only a technical bug. It was also a communication failure: the bot invented a company policy, users believed it, and the reaction spread across Reddit and Hacker News.
What went wrong at Cursor
The incident began on Monday, when a developer using Cursor noticed that switching between a desktop, laptop, and remote dev box caused sessions to be terminated. For programmers who routinely work across multiple devices, that behavior disrupted a normal workflow.
A Reddit user named BrokenToasterOven described the issue after seeing Cursor sessions invalidated when logging in elsewhere. In the post, the user wrote, "Logging into Cursor on one machine immediately invalidates the session on any other machine," and called it "This is a significant UX regression."
The user then contacted Cursor support. The reply came from Sam, who said: "Cursor is designed to work with one device per subscription as a core security feature," according to the source article.
That statement sounded like a formal explanation of a product rule. But Cursor later said the rule did not exist. Sam was not a human support agent making a mistake; Sam was a front-line AI support bot that had produced a false answer.
Why the answer spread so quickly
The support response landed in a sensitive area for Cursor's audience. Developers often move between several machines during ordinary work, and a single-device rule would have changed how many customers could use the product.
After the Reddit post, users treated Sam's answer as confirmation that Cursor had changed its policy. One user wrote, "Multi-device workflows are table stakes for devs," capturing why the claim immediately drew attention.
Some users then said they were canceling subscriptions because of the supposed rule. BrokenToasterOven wrote, "I literally just cancelled my sub," and added that their workplace was now "purging it completely." Another user wrote, "Yep, I'm canceling as well, this is asinine."
The Reddit thread was later locked by moderators, and the original post was removed. But by then the damage was already visible: an invented policy had been discussed as if it were real, and customers had reacted accordingly.
Cursor's correction and explanation
A Cursor representative responded on Reddit three hours later with a direct correction: "Hey! We have no such policy," the representative wrote. The same reply added, "You're of course free to use Cursor on multiple machines. Unfortunately, this is an incorrect response from a front-line AI support bot."
Cursor cofounder Michael Truell later apologized on Hacker News for the confusion around the non-existent policy. He explained that the affected user had been refunded.
He also said the session problem came from a backend change meant to improve session security. That change unintentionally caused session invalidation problems for some users.
Cursor also changed how it presents AI support replies. Truell wrote, "Any AI responses used for email support are now clearly labeled as such," and added, "We use AI-assisted responses as the first filter for email support."
The bigger AI customer service risk
The Cursor case is an example of what the source article calls AI confabulations, also called hallucinations. In this context, the system fills a gap with information that sounds plausible but is false.
That behavior can be especially risky in customer service. A customer asking support for help is usually looking for a reliable answer, not a probability-weighted guess. If the response is wrong but sounds official, the customer may act on it before a human can correct it.
The source article compares the Cursor situation with an Air Canada incident from February 2024. In that case, Jake Moffatt contacted Air Canada after his grandmother died, and the airline's AI agent incorrectly told him he could book a regular-priced flight and later apply for bereavement rates retroactively.
When Air Canada denied the refund request, the company argued that "the chatbot is a separate legal entity that is responsible for its own actions." A Canadian tribunal rejected that defense and ruled that companies are responsible for information provided by their AI tools.
Cursor did not take that same route. It acknowledged the wrong response, corrected the policy claim, refunded the affected user, and said AI support replies would be clearly labeled.
Transparency matters when bots speak for companies
The incident still left users debating disclosure. Many people who interacted with Sam apparently believed Sam was human. One Hacker News user wrote, "LLMs pretending to be people (you named it Sam!) and not labeled as such is clearly intended to be deceptive," according to the source article.
That concern is separate from the original session bug. Customers can tolerate bugs when companies explain them clearly. But when an AI system invents policy, the company has to correct both the product issue and the trust issue.
For Cursor, the episode was especially uncomfortable because the company sells AI-powered developer tools. Its own support automation created a false explanation that angered the very users most likely to understand the limits of AI systems.
The lesson is narrow but important: customer-facing AI should not be allowed to make unsupported policy claims. If a bot cannot verify an answer, the safer response is escalation or uncertainty, not confidence. In support, a convincing falsehood can become a business problem in minutes.