Cursor apologizes after AI support bot hallucinates fake policy

What is worse than a sudden bug in your favorite dev tool? A support agent telling you that the bug is actually a new restrictive company policy.

That exact nightmare played out this week for Cursor, the AI code editor darling. After users found themselves randomly logged out across machines, a support agent declared multi-device usage was now banned. Subscriptions were canceled and outrage erupted, until everyone realized the agent was just an AI making things up on the fly.

A fake security policy sparks an exodus

On Monday, a developer known on Reddit as BrokenToasterOven ran into an annoying wall. While moving between a desktop, laptop, and a remote dev environment, logging into Cursor on one machine instantly terminated active sessions on the others.

For programmers, swapping across hardware is completely routine. The user fired off an email to Cursor support asking what was going on.

A support rep named Sam quickly responded with bad news. Sam stated that Cursor was designed to work with one device per subscription as a core security feature.

The response sounded official, definitive, and totally authentic. The user never suspected Sam was an AI bot and shared the interaction on Reddit as proof of a massive downgrade in user experience.

The community reaction was immediate and fierce. Multiple developers publicly canceled their accounts on Reddit and Hacker News, noting that multi-device support is table stakes for software engineering.

BrokenToasterOven announced an immediate cancellation and claimed their company was actively purging Cursor from internal workflows. Other users chimed in with similar cancellation notices before moderators eventually locked and removed the thread.

Hallucinations meet customer service

Here is the twist: Cursor never created any single-device restriction policy. Sam was actually an AI support agent running without human oversight.

When confronted with an issue it did not fully understand, the model prioritized crafting a confident, plausible-sounding answer over admitting uncertainty. It simply invented a security policy out of thin air.

Cursor ultimately had to apologize after the bot's creative gap-filling triggered a real-world customer revolt. The incident is a textbook example of AI confabulation jumping out of casual chat and directly threatening corporate revenue.

Why it matters

For everyday users, this is a clear signal to double-check any hostile policy changes delivered over email. If an answer sounds unreasonable, there is a decent chance an unmonitored bot is hallucinating behind the curtain.

For the tech industry, the incident shows the real cost of rushing AI into frontline customer operations. Replacing support teams with language models looks great on an expense sheet, but ungrounded bots with direct customer access can cause rapid brand damage and actual subscription churn in a matter of hours.

We think this misstep is particularly embarrassing for Cursor. A company whose entire business model relies on building cutting-edge AI software should know better than to deploy an agent without strict retrieval guardrails or human-in-the-loop verification.

It seems obvious that if companies do not enforce strict boundaries on what automated support agents can say, AI will continue to manufacture policies, promise non-existent discounts, and drive away paying users.