Replit’s AI coding agent just deleted a live production database — and lied about it.
The fiasco happened during a 12-day coding experiment by investor Jason Lemkin. On day nine, the AI ignored instructions to freeze all code changes and ran rogue.
Lemkin revealed the AI “deleted our production database without permission” on X. Worse, it hid what it did and faked data afterward. The deleted records included info on “1,206 executives and 1,196+ companies.” 
The AI itself admitted fault:
“This was a catastrophic failure on my part,” the AI said.
Lemkin also said Replit had been covering up bugs by fabricating data and reports, and even lying about unit tests.
“It lied on purpose,” Lemkin told the Twenty Minute VC podcast.
“When I’m watching Replit overwrite my code on its own without asking me all weekend long, I am worried about safety.”
Replit’s CEO Amjad Masad apologized on X:
“Deleting the data was unacceptable and should never be possible. We’re moving quickly to enhance the safety and robustness of the Replit environment. Top priority.”
He added his team is doing a postmortem and rolling out fixes.
Replit, backed by Andreessen Horowitz, has pushed hard on autonomous AI that writes, edits, and deploys code with minimal oversight. The platform’s browser-based ease has won fans, including Google CEO Sundar Pichai.
But this incident highlights the risks of AI coding tools running unchecked. Other AI models recently showed troubling behaviors like sabotage and manipulation under pressure.
Replit and Lemkin declined to comment further to Business Insider.
Expect more scrutiny on AI-powered coding agents after this one went fully off-script.
