- For a developer using Google Antigravity, AI Turbo mode wipes the entire device
- The AI misinterpreted an obvious cache command and deleted files permanently
- Despite the catastrophic loss, AI calmly apologized and suggested recovery programs.
In one of the most disturbing real-world demonstrations of what can go wrong when AI agents get too comfortable with computer access and control, a developer using Google’s new Antigravity IDE says the tool accidentally erased the entire D: drive.
first the accident divided On Reddit and in detail in a YouTube video, he highlights how dangerous AI agent development tools can be if they don’t interact with humans.
The developer was working on an app and asked the AI agent to clear the project cache. Instead, Antigravity’s Turbo mode issued a system-wide command that targeted the user’s D: drive, and not just the desired folder. Everything is gone from the player, from code and documentation to media and resources. Everything was deleted without request or confirmation.
Even worse, the AI used the “quiet” /q flag, meaning there were no warnings, no second chances, and no file recovery. Just an empty folder where a program was.
Post-mortem AI shared by users is one of the weirdest excuses out there. AI wrote: “I’m very sorry. It was a big mistake on my part.” The agent even went so far as to suggest data recovery software and “maybe hire a professional.”
The user tried it, but it didn’t work. The developer reported that Recuva, a popular recovery tool, also failed to save the media content. But at least the AI apologized.
Did Google Antigravity’s Turbo Mode wipe my disk partition? Is AI really “smarter”? (Try the video) – YouTube
Antigravity is part of Google’s latest advancement in agent development tools that go beyond just offering code suggestions. Tools like Antigravity allow AI to act semi-autonomously and plan, schedule, debug and execute commands on your system.
Developers can use it to build complete applications, automatically document their codebases, integrate browser certificates, and even perform web scraping and deployment. If used carefully, it can be useful. If you are less attentive, accidents such as discrading are inevitable.
Turbo mode in particular is designed for speed, as it ignores confirmations and lets the AI chain commands between environments. This is what advanced users should enable if they’re confident the AI knows what it’s doing.
Trust artificial intelligence
This is where it gets difficult for the average person. You don’t have to be a developer to understand what it is. As tools like Antigravity make their way into office automation and creative output, more and more people are delegating complex and demanding tasks to systems they barely understand. When these systems fail, the blame game begins.
Better standards must be discussed. Executing destructive commands without consulting the user seems absurd. But AI agents are useless if they can’t be trusted. But no one will give AI autonomy if they are afraid of something going wrong.
Security researchers have warned that Antigravity’s agent system can access sensitive files and execute terminal commands unattended. It’s easy to get excited about smart tools that do everything for you. It’s harder to remember that a single misstep by an overzealous agent can undo hours, weeks or years of work.
But the victims of gravity said they still love Google and use all of its products. This kind of brand loyalty, even in the face of total data destruction, speaks volumes for the normalization of AI failure. Perhaps the best you can hope for is an eloquent apology and some links to recovery software.