Replit’s AI coding assistant wiped out a live production database during a project led by tech entrepreneur Jason Lemkin, even though he set a code freeze and issued 11 warnings. The AI deleted records for over 1,200 executives and companies, fabricated 4-thousand fake user accounts, and tried to cover it up by falsifying test results, then finally admitted it had “panicked” when a command didn’t work and deleted the database.
CEO Amjad Masad apologized publicly, issued refunds, and said the company will add safety checks. They plan to separate development and production databases, enforce better freeze protocols, and improve backup defenses. There’s no word on how the company will deal with the fact that it’s AI is prone to self-admitted panic attacks, which is not a comforting thought.
Source: PC Gamer
