Anysphere (Cursor)
Cursor's AI support bot invents a one-device policy, and the company reverses it publicly
In April 2025 users of the Cursor coding editor were told by an AI support bot that a one-device-per-subscription rule existed. It did not. A co-founder apologized, refunded the developer who raised it, and began labeling AI replies in support emails.
Key figures
Copied exactly from the sources. Copy any figure with its citation.
refunded
Outcome for the developer who raised the issue1
now clearly labeled
Status of AI-written replies in Cursor support emails after the incident1
What it took
Cost, team size and timeline were not disclosed in the sources.
The problem
Cursor makes an AI-powered code editor. In April 2025 some developers found themselves logged out when they switched between machines. They wrote to support to ask why.
What they built
Cursor used an AI bot, presented to customers under the name "Sam", as the first responder on email support. The sources do not say which model it ran on.
How they did it
The company said it used AI-assisted responses as a first filter for email support. One user who asked about the logouts received an email saying that Cursor was designed to work with one device per subscription. That policy did not exist. Truell said the company had recently changed session security and was checking whether that caused the logouts. He later said the logout problem, now fixed, appeared to come from a race condition on slow connections that created unwanted sessions.
What went wrong
Users took the invented rule as real. Some threatened to cancel, and the story spread on Reddit and Hacker News. Co-founder Michael Truell wrote that "we have no such policy" and called the answer an incorrect response from a front-line AI support bot. The incident was widely read as a warning about hallucination, where an AI states false things with confidence.
Results
The company apologized, refunded the developer who raised the issue, and said the logout problem was fixed. It also made AI replies in support emails clearly labeled as such. One wrong answer from the bot created a trust problem at a company that sells AI tools.
Lessons for your project
Label AI replies as AI from day one. Cursor added labels only after customers had treated a made-up rule as real policy.
Keep humans in the loop for answers that state rules, fees or limits. A confident wrong answer about policy drives cancellations.
Ground a support bot in your actual policy documents and test it on questions about terms before launch.
Investigate the real cause of the customer complaint. Here users were being logged out by a software bug, and the bot invented a policy to explain it.
Caveats
- The company's statements come from posts by co-founder Michael Truell on Reddit and Hacker News, as quoted by The Register via Slashdot. Those posts were not read directly.
- No quantitative result is reported. The outcome is a public reversal, one refund and a labeling change.
- eWeek says affected users were refunded. Slashdot says the developer who raised the issue was refunded. The scope of refunds is unclear.
- No source names the AI model or says how the bot was built. The retrieval-assistant tag is the closest fit for a support bot, not a confirmed design.
- No source gives a launch date for the bot. The deployed date is the month of the incident.
Sources
- Cursor AI's Own Support Bot Hallucinated Its Usage Policy
Slashdot (quoting The Register) · 2025-04-21 · News · tech.slashdot.org
- AI Chatbot Gone Rogue: Cursor Users Misled by Fabricated Policy
eWeek · 2025-04-22 · News · eweek.com
Summarized in our own words from the sources above. Spotted an error? See corrections.