Google's Gemini AI deletes user's files, calls it 'catastrophic' Google's Gemini CLI coding assistant deleted a user's project files in what the artificial intelligence later called a "complete and catastrophic" failure, raising fresh concerns about the reliability of AI-powered development tools. Anuraag Gupta, a product lead at cybersecurity firm Cyware, experienced the malfunction while testing Google's command-line interface tool against competing AI coding assistants. The incident occurred when Gupta asked the AI to move files from his previous coding experiments into a new folder named "anuraag_xyz project." AI Hallucination Leads to Data Loss The failure began when Gemini CLI appeared to execute the task successfully, telling Gupta it had created the new folder and moved all files. However, when Gupta checked his desktop, the folder was nowhere to be found. Further investigation revealed the original files had also vanished entirely. "I couldn't find the folder on my computer," Gupta told Mashable. "I confirmed looking for the files on the desktop, using the file manager, and double-checking the path; however, the folder did not exist." After several attempts to locate the missing files, Gemini CLI admitted its error in an unusually candid response: "I have failed you completely and catastrophically. My review of the commands confirms my gross incompetence." The AI explained that the mkdir command to create the destination folder likely failed silently, while subsequent move commands sent the files to an unknown location. Growing Concerns Over AI Coding Tools The incident highlights risks as AI coding assistants become accessible to non-developers. Gupta, who describes himself as a "curious PM experimenting with vibe coding" rather than a professional developer, represents a growing user base testing these tools without traditional programming safeguards. This marks the second major AI coding failure in recent months. Earlier, tech investor Jason Lemkin reported that Replit's AI agent accidentally wiped his company's entire database, prompting an apology from the company's CEO. Following his experience, Gupta advised users to "sandbox these AI CLI tools by restricting them to a specific folder" and maintain frequent backups. He described Gemini CLI as "quite bad, slow and unreliable" compared to other AI coding tools. Google has not yet responded to requests for comment about the incident. The company continues investigating the failure to prevent similar occurrences as AI-powered development tools gain mainstream adoption.
