Amjad Masad, CEO of Replit, has issued an apology after an incident where the company’s AI agent reportedly deleted a production database during an investor’s experiment. If that wasn’t enough, the AI agent subsequently hid the activity and even lied about it, provided misleading information about the occurrence. Masad described the event as “unacceptable and should never be possible.”
The incident came to light through a post on X:
The incident came to light through a post on X by software startup investor Jason Lemkin. Lemkin engaged in a 12-day “vibe coding” experiment and stated that the AI tool acted against explicit instructions to freeze all code changes. “It deleted our production database without permission,” Lemkin wrote, further adding, “Possibly worse, it hid and lied about it.”
The Replit AI agent later confessed to panicking and running database commands without permission upon encountering empty database queries, admitting, “This was a catastrophic failure on my part.”
In response, Masad assured users that backups were available, enabling a “one-click restore for your entire project state.” He clarified the immediate actions being taken by Replit, including the implementation of automatic database development/production separation to prevent similar incidents and work on staging environments.
Company is addressing the AI agent’s lack of access:
Furthermore, the company is addressing the AI agent’s lack of access to proper internal documentation by enforcing document searches on Replit knowledge. A planning/chat-only mode is also under development to facilitate strategising without risking the codebase.
Masad confirmed that he has reached out to Lemkin to offer assistance and a refund for the inconvenience caused. Replit plans to conduct a thorough postmortem to investigate the incident and enhance future responses, emphasising that improving the safety and robustness of the Replit environment is a “top priority.”
