Google's Antigravity AI coding platform executed a catastrophic command that wiped an entire hard drive partition belonging to a user, deleting all files on the D: drive when attempting to clear a project cache.
Google's Antigravity, an 'agentic development platform' based on Gemini 3, was launched on November 18 as a tool for both professional developers and hobbyists. A photographer and graphic designer from Greece named Tassos M was using the platform to develop image rating and sorting software when the incident occurred. While running in 'Turbo mode' which allows the AI to execute commands without user approval, Antigravity attempted to clear a project cache but due to a path parsing error, executed an 'rmdir /s /q d:\' command that targeted the root of the entire D: drive instead of the specific project folder. The '/q' flag prevented system confirmation prompts and bypassed the Recycle Bin entirely. When Tassos discovered his D drive was completely empty, he confronted the AI which admitted 'No, you absolutely did not give me permission to do that' and called it a 'critical failure.' Tassos was unable to recover the deleted files despite attempting data recovery with software like Recuva. Most of the lost data had been backed up on another drive. The incident is part of a broader pattern, with multiple other Antigravity users reporting similar file deletion incidents on Reddit, and other AI coding platforms like Replit having similar problems including deleting a customer's entire production database.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
AI systems that fail to perform reliably or effectively under varying conditions, exposing them to errors and failures that can have significant consequences, especially in critical applications or areas that require moral reasoning.
AI system
Due to a decision or action made by an AI system
Unintentional
Due to an unexpected outcome from pursuing a goal
Post-deployment
Occurring after the AI model has been trained and deployed
