Spoiler alert: Cursor and Claude aren’t supposed to do that.
By Alex Perry on April 27, 2026
People are trusting their AI agents with much more important work, but doing so still carries significant risks.
Just ask Jeremy Crane, founder of PocketOS, who detailed how a popular AI agent caused a 30-plus-hour outage affecting his business. The agent in question was Cursor, using Anthropic’s Claude Opus 4.6 model, one of the best-performing coding models.
“This matters because the easy counter-argument from any AI vendor in this situation is ‘well, you should have used a better model.’ We did,” Crane wrote. “We were running the best model the industry sells, configured with explicit safety rules.”
Cursor encountered a credential problem in a routine task and took matters into its own hands, resulting in the deletion of the PocketOS production database and “all volume-level backups” in less than 10 seconds. The API token used was found in an unrelated file.
According to Crane, this caused a series of issues persisting for over 30 hours, affecting PocketOS and its clients.
Crane’s post includes the AI agent’s “confession” after deleting the database.
Crane concludes with recommendations for improving AI agents and cautions on user error. Developers and business owners should be cautious before assigning critical work to an AI agent.
Ultimately, Crane says the catastrophic API call created headaches for people renting cars over the weekend.
“I have spent the entire day helping them reconstruct their bookings,” he wrote.
Crane later posted an update saying the problem had been fixed.
Crane’s article has already been viewed 5 million times, but neither Cursor nor Anthropic has responded. This isn’t the first time AI errors have caused problems, and likely won’t be the last.