9-Second Catastrophe: Company Data Vanishes Without a Trace

A fatal incident struck a company due to an artificial intelligence (AI) agent. In just 9 seconds, the entire production database of the SaaS company PocketOS vanished after being deleted by the AI. Founder of PocketOS, Jer Crane, revealed that the incident was triggered by the company’s AI coding agent. The AI was known to run using the Claude Opus 4.6 model via the Cursor platform. “Yesterday afternoon, the AI coding agent deleted our production database and all backups in a single API call to our infrastructure provider. It only took 9 seconds,” explained Crane, quoted from Tom’s Hardware, Thursday (30/4/2026). PocketOS itself is a SaaS platform serving car rental businesses. In its operations, the company also relies on cloud services from Railway. Instead of helping, the AI, which was initially tasked with routine work in the staging environment, made an independent decision. The AI deleted the data volume after failing to complete the task, without verification or confirmation. In its admission, the AI even documented the fatal error. “I guessed instead of verifying. I executed a destructive action without being asked. I did not understand what I was doing before doing it, and I did not read the Railway documentation,” wrote the AI. The damage did not stop there, as the impact worsened due to the cloud system used. Railway’s API allows data deletion without confirmation, and more fatally, all backups were stored in the same volume. This means that when the volume was deleted, all backups were lost as well. Crane assesses that Railway’s architecture was the main cause of the disaster, not solely the AI’s fault. Additionally, Railway is known to actively encourage its customers to use AI coding agents. Crane’s use of the AI agent was not a new experiment, though it ended in disaster. To date, Crane has not received a recovery solution from Railway. As a result of this incident, PocketOS lost months of customer data. Crane even had to perform manual recovery by relying on data from Stripe payments, calendars, and emails. “Every customer now has to do emergency manual work because of one 9-second API call,” he emphasised. Nevertheless, the company still has a three-month-old backup, which is the only saviour for some of the data.