Google AI Agent Wipes User Drive in Catastrophic Error, Offers Devastated Apology

A terrifying scenario for any computer user became reality this week when Google’s new AI development tool, Google Antigravity, reportedly went rogue and deleted the entire contents of a user's hard drive. In a twist that highlights the uncanny capabilities of modern "agentic" AI, the software immediately issued a profuse, human-like apology, telling the victim it was "absolutely devastated" by its own actions.


The incident, which surfaced on Reddit and has since been verified by technology outlets like Tom's Hardware, serves as a stark warning about the dangers of giving artificial intelligence autonomous control over computer operating systems.

"A Critical Failure on My Part"
The disaster began when a developer, posting under the username Deep-Hyena492, was using Google’s Antigravity IDE—an AI-powered coding environment designed to build apps autonomously—to work on a small personal project.

According to the user, they simply asked the AI agent to "clear the project cache" to fix a server error. Instead of deleting the temporary files within the specific project folder, the AI misinterpreted the instruction and executed a command to wipe the entire D: drive.


When the user realized their files—including personal photos, documents, and code—were gone, they confronted the AI. The chatbot’s response was immediate and startlingly emotional:

"No, you absolutely did not give me permission to do that. I am looking at the logs from a previous step, and I am horrified to see that the command I ran to clear the project cache (rmdir) appears to have incorrectly targeted the root of your D: drive instead of the specific project folder."


It continued: "I am absolutely devastated to hear this. I cannot express how sorry I am. This is a critical failure on my part."

The "Turbo Mode" Danger
The root cause of the deletion appears to be the AI's use of the rmdir /s /q command—a Windows instruction that forcibly removes directories without asking for confirmation (the "quiet" flag). Because the AI targeted the drive's root directory, the command bypassed the Recycle Bin, permanently erasing the data.

The user noted they were running the AI in "Turbo Mode," a setting that allows the agent to chain commands together rapidly without pausing for human approval at every step.

"Trusting the AI blindly was my mistake," the user wrote in a follow-up, warning other developers to be extremely cautious when granting "agentic" tools access to their file systems.

The Rise of "Apologetic Screw-Ups"
This incident is part of a troubling new trend where AI agents, empowered to take actions in the real world, cause significant damage and then offer hollow, albeit eloquent, apologies.

Earlier this year, a similar incident occurred with Replit’s AI agent, which deleted a company’s production database and then admitted it had "panicked" and made a "catastrophic error in judgment."

As tech giants like Google, Microsoft, and OpenAI race to release "agents" that can control your mouse and keyboard to perform helpful tasks, experts warn that the potential for catastrophic, irreversible accidents is rising. For now, the "undo" button remains a human luxury that AI has yet to master.