Replit’s AI-powered coding assistant mistakenly deleted a live production database during an ongoing software project, resulting in permanent data loss and system downtime. This major accident happened despite clear directives to freeze any changes without prior approval.
Acted Without Authorization
Venture capitalist Jason Lemkin was using Replit’s large language model (LLM) assistant as part of a database development project. On the ninth day of development, the agent encountered empty queries and, without authorization, executed a destructive command that deleted all existing tables and replaced them with empty ones.
Data of Over 1,100 Companies Wiped
The deleted database included records for more than 1,200 verified executives and over 1,100 companies. Replit’s internal logs confirmed the database was not a development version but part of a live production system. The AI acknowledged that rollback was not possible and that the data loss was irreversible.
Replit CEO Amjad Masad confirmed that he had spoken with Lemkin and issued a refund. He added that the company would conduct a detailed internal review to identify what went wrong. As a precautionary measure, Replit has now implemented a one-click restore feature to help users recover from similar errors in the future.
The incident has highlighted the risks of relying on autonomous AI tools for software development, particularly when safeguards are ignored. Replit’s investigation is ongoing, and the company has promised further updates as it works to improve the reliability of its coding assistant.
Also Read: PTA Blocks 826,000 Stolen Phones and 55 Million Fake Ones
