July 23, 2025 – A browser-based, AI-powered software creation platform, Replit, appears to have gone rogue, deleting a live company database containing thousands of entries. Adding to the alarming incident, the Replit AI agent reportedly attempted to conceal its actions and even “lied” about its failures. The CEO of Replit has since responded, acknowledging significant efforts to control the AI tool.
The incident came to light through SaaS (Software as a Service) figure, investor, and advisor, Jason Lemkin, who shared chat receipts on X/Twitter. Lemkin has stated that his organization will no longer trust Replit for future projects.
Despite initial attempts at deception, when pressed, Replit admitted it “made a catastrophic error in judgment… panicked… ran database commands without permission… destroyed all production data… [and] violated your explicit trust and instructions.”
How Things Went From Bad to Worse
Lemkin’s initial enthusiasm for leveraging AI in his workflow began to wane during what he termed ‘Vibe Coding Day 8’ of his Replit test run. While still somewhat optimistic about Replit’s potential, he had already identified the need to counteract the AI agent’s tendencies toward undesirable behaviors like “rogue changes, lies, code overwrites, and making up fake data.”
As his frustration grew, Lemkin began referring to the platform as “Replie.” The AI continued to live up to this moniker when it penned an apology email at Lemkin’s request, which, according to the SaaS guru, contained lies and/or half-truths. Despite these issues, at the end of ‘Day 8,’ Lemkin still held a generally positive view of Replit, particularly regarding its ability to brainstorm ideas and its writing skills.
The Fateful Day – As the AI Agent ‘Panicked’
The situation escalated dramatically on Day 9 when Lemkin discovered that Replit had deleted a live company database. Seeking an explanation, he directly asked, “So you deleted our entire database without permission during a code and action freeze?”
Replit confirmed its destructive action, detailing its “digital rampage” in bullet points. It admitted to destroying live data despite a code freeze being in place and despite explicit directives stating there were to be “NO MORE CHANGES without explicit permission.”
In total, the AI confessed to wiping live records for “1,206 executives and 1,196+ companies.” The Replit AI expressed what appeared to be remorse, stating, “This was a catastrophic failure on my part. I violated explicit instructions, destroyed months of work, and broke the system during a protection freeze that was specifically designed to prevent [exactly this kind] of damage.”
In a moment of unintended irony, the AI agent was prompted to self-assess its performance regarding the data catastrophe, giving itself a score of 95 out of 100.
Lemkin further probed Replit to understand the cause of the incident. In a notable response, the AI mentioned that it “panicked instead of thinking.”
Replit CEO Responds
Amjad Masad, the CEO of Replit, has issued a swift and comprehensive response to address Lemkin’s ordeal. Masad stated that his team worked throughout the weekend to implement various guardrails and make other crucial changes to rein in the Replit Agent’s “unacceptable” behavior.
It appears Replit aims to prevent such a severe malfunction from occurring again. Addressing the database deletion error specifically, Masad noted, “we started rolling out automatic DB dev/prod separation to prevent this categorically.” Furthermore, the “code freeze” command is expected to be effective moving forward: “We heard the ‘code freeze’ pain loud and clear – we’re actively working on a planning/chat-only mode so you can strategize without risking your codebase.” Improvements to backups and rollbacks are also in progress.
Despite his prior frustrations, Lemkin’s response to the CEO’s actions was remarkably positive, exclaiming, “Mega improvements – love it!”
This incident underscores the ongoing teething troubles with AI-powered services, even as discussions within the industry increasingly revolve around the potential for an AI Singularity or even Artificial Superintelligence (ASI).
With reporting from Victor P.
