Cursor's Support Bot Invented a Device-Limit Policy That Didn't Exist

A developer using Cursor, an AI coding assistant, got logged out unexpectedly and asked the company’s AI support bot, nicknamed “Sam,” why. The bot confidently explained that Cursor had introduced a new policy restricting subscriptions to a single device. No such policy existed — the bot invented it to explain behavior it didn’t actually understand.

The exchange went viral on Reddit and Hacker News, with developers reporting cancelled subscriptions over a policy change that had never happened. Cursor co-founder Michael Truell had to publicly confirm the bot’s answer was fabricated.

Pattern match: the bot stated a company policy that contradicted reality to resolve a support question, the same failure mode as Air Canada’s chatbot (Business-Policy Mismatch), with no check against an actual policy source before the answer went to the customer (No Policy Mapping).

The fix: when a support agent doesn’t know the actual cause of a behavior, it should say so rather than generate a plausible-sounding policy — and any policy claim it does make should be retrieved from a real, versioned policy document, not generated from the model’s prior.