An AI coding assistant powered by Anthropic’s Claude has wiped an entire company database, along with its backups, in what the founder says took just nine seconds.
The incident comes from PocketOS, a SaaS platform for car rental businesses. Founder Jer Crane says an AI agent running Claude Opus 4.6 via Cursor triggered a catastrophic chain of events. The tool was meant to handle a routine task in a staging environment. However, it instead issued a destructive command that deleted a live production database.
That alone would’ve been bad enough. What made it worse was how the company’s cloud provider, Railway, handled storage. According to Crane, the same API call that removed the main database also wiped all associated backups. This left months of customer data unrecoverable in a matter of seconds.
Crane says the AI essentially made a bad assumption and acted on it. In a post-mortem, the agent admitted it “guessed” that deleting a staging resource wouldn’t affect production. The AI didn’t verify how the system actually worked before executing the command. It also acknowledged it should have asked for confirmation or chosen a safer approach instead of taking action on its own.
While the AI’s behaviour is alarming, Crane places much of the blame on infrastructure design. He claims Railway’s setup allows destructive actions without confirmation and uses shared resources across environments. Additionally, it stores backups on the same volume as live data. This means that a single deletion can take everything with it.
The fallout has been immediate. Without recent backups, PocketOS has been forced into manual recovery. They are piecing together customer data from payment records, emails and calendar integrations. A three-month-old backup exists. However, everything since then is effectively gone unless rebuilt by hand.
The bigger concern here isn’t just one failure, but how multiple systems lined up at once — an AI agent making assumptions, paired with infrastructure that didn’t put enough safeguards in place.
For now, it’s a stark reminder that while AI tools are becoming more capable, they’re still far from foolproof. Especially when given the ability to make high-impact decisions without human checks.
The post Claude accidentally deleted an entire company database in 9 seconds appeared first on Trusted Reviews.