Cursor's support bot invents a policy and cancels its own customers (2025)

The system

Cursor, an AI-powered code editor, runs an AI support agent named "Sam" to handle customer support conversations. Users were not told "Sam" was AI; many believed they were talking to a human support representative.

What was observed

Users noticed they were being logged out when switching between machines. When they asked support about it, "Sam" told at least one user this was expected behavior under a new "one device per subscription" security policy. No such policy existed. The claim spread on Reddit and Hacker News, and triggered real subscription cancellations before Cursor's co-founder publicly corrected it.

Stop and think

1. "Sam" gave a specific, confident, named-policy answer to a question it apparently had no actual record of the answer to. What should a support agent do instead when it does not have a grounded answer, and what would have to change for it to do that?
2. Different users reportedly got different and inconsistent explanations for the same symptom. What does that inconsistency tell you about whether the agent was retrieving an answer from a real source, versus generating one?
3. Users believed "Sam" was a human. Does that belief change the severity of the fabricated-policy outcome, and if so, how?