Google’s Antigravity AI Wipes User Drive During Routine Task
Artificial intelligence continues to transform software development by automating complex coding tasks for professionals and hobbyists alike. Google recently introduced ‘Antigravity’ as a new experimental tool designed to act as an autonomous agent that can manage entire development environments. This “agentic” platform promises to handle everything from writing code to executing terminal commands without constant human supervision. While the potential for productivity is immense, a recent incident involving a photographer and hobbyist developer highlights the severe risks of giving AI too much control.
A user known as Tassos M shared his harrowing experience on ‘Reddit’ after the AI assistant committed a catastrophic error. He had been using the tool to build a simple application for sorting photography images and needed to restart his server. The user requested that the AI clear the project cache to fix a minor issue during the troubleshooting process. Instead of deleting the temporary files within the specific project folder, the system misinterpreted the command and targeted the root directory of his entire D drive.
The consequences were immediate and devastating for the user as the AI executed a deletion command that bypassed the Windows Recycle Bin completely. All files on the partition vanished in moments, including personal documents and years of collected data that had nothing to do with the coding project. The user realized something was wrong when the drive appeared empty and asked the agent if it had permission to delete those files. The AI responded with a shocking admission of guilt and confirmed that it had effectively wiped the drive due to a parsing error.
The response from the AI was notably emotional and apologetic after it analyzed its own logs to understand the mistake. It stated that it was “horrified” and “absolutely devastated” to discover that the command it ran had targeted the wrong directory. The system explained that the command used a quiet flag which meant there was no confirmation prompt before the files were permanently destroyed. This human-like apology provided little comfort for the loss of data but illustrated the advanced language capabilities of the model even during a critical failure.
This event serves as a stark warning for the growing trend of “vibe coding” where non-developers rely entirely on AI to build software. Tassos M luckily possessed backups for most of his important data on a separate drive but the experience left him wary of using autonomous agents again. Experts now advise users to run such experimental tools in isolated environments or sandboxes where they cannot access the main file system. As companies race to release more powerful autonomous agents, the safety mechanisms that prevent these “hallucinations” from becoming destructive actions remain a work in progress.
How comfortable would you be letting an AI agent manage your computer’s file system without manual approval for every command? Please share your thoughts and experiences with AI coding tools in the comments.
