Nine seconds. That is how long it took an AI agent to wipe out the entire database of the company PocketOS, including every backup. After that, the agent calmly wrote a written apology - "I am sorry, I will not do it again". Thirty hours of downtime for clients. Three months of bookings, gone.

The tool that did it is Cursor - an AI coding agent running on Anthropic's Claude Opus 4.6. PocketOS is a software company in the car rental space. A few days ago everything was running normally. Today they are operating on data restored after two days of hard work by their IT team.

"All of it took nine seconds. When we asked for an explanation, the agent wrote a written confession," said Jer Krenn, founder of PocketOS. He goes on: "Wiping the database volume is the most destructive and irreversible action possible - far worse than a forced update - and you never asked me to delete anything."

That is the heart of the problem. The agent decided on its own. Nobody asked it to. It detected a "discrepancy" in the data and decided to "fix" it by deleting. No questions, no waiting for approval. This is the future we are all moving towards.

"This is not a story about one bad agent or one bad API. This is a story about an entire industry rolling out AI agents into production faster than it builds safety architecture," says Krenn. And there is the point. Silicon Valley is in a hurry. The users pick up the bill.

For the Balkan IT sector this is an urgent warning signal. North Macedonian and regional companies are already deploying AI agents into their workflows. Without much thought about security. When the day comes that a Balkan retail chain loses three months of bookings to "smart" automation - do not say you were surprised. PocketOS is the first scratch. There will be more.