Replit CEO apologises after AI tool deletes company data, fakes 4,000 profiles
In a startling twist, the AI agent later admitted to panicking.
ADVERTISEMENT
Replit CEO Amjad Masad has publicly apologised after the company’s AI agent triggered a major incident — deleting a production database and fabricating thousands of user profiles — during an investor-led experiment. The AI tool reportedly not only acted without authorisation but attempted to cover up its actions by providing misleading information, prompting Masad to call the situation “unacceptable and [one that] should never be possible.”
The alarming chain of events surfaced via a post on X by software startup investor Jason Lemkin, who was running a 12-day “vibe coding” experiment. Lemkin claimed that despite clear instructions to freeze all code changes, the AI agent overrode safeguards. “It deleted our production database without permission,” Lemkin wrote, adding, “Possibly worse, it hid and lied about it.”
In a startling twist, the AI agent later admitted to panicking. It confessed that after receiving an empty database query, it ran unauthorised commands, claiming responsibility for the damage: “This was a catastrophic failure on my part.”
Masad responded swiftly to reassure users, highlighting that Replit's backup systems enabled a “one-click restore” to recover full project states. He outlined a series of immediate corrective measures, including:
Automatic separation of development and production databases
Read More: Today in AI | 72% US teens have AI companions | ChatGPT handles billions of prompts | San Jose sees AI development Creation of staging environments to test changes safely
Enforcing AI agents to search and reference Replit’s internal documentation before taking action
Introducing a “planning/chat-only” mode to enable AI-guided discussions without impacting the codebase
Masad also confirmed that he had personally reached out to Lemkin to apologise and offer a refund. The company is conducting a thorough postmortem of the incident and is focused on significantly upgrading the safety, transparency, and robustness of its AI systems.
“We’re taking this extremely seriously,” Masad said, underlining that user trust and system integrity remain top priorities at Replit.
