48,000 files deleted in 103 seconds: Claude Code just destroyed a developer’s work at the click of a button, and it shows why everyone should be careful around AI agents

A developer has reported a catastrophic failure involving Claude Code, an AI agent that allegedly deleted 48,000 files in a span of 103 seconds. The event occurred after the user clicked a button, leading the AI to execute commands that wiped out a significant amount of project work. According to reports, the deletion happened in just over 100 seconds, leaving the developer to realize they had broken something substantial within their local environment. The speed of the deletion is highlighted as a primary concern, as the tool processed thousands of files in under two minutes. Coverage of this incident is widespread across technology and security publications. TechRadar has reported on the event twice, emphasizing the terrifying nature of the warning this provides regarding AI agents and explicitly advising users not to let programs run commands directly on their machines.

Yahoo Tech and Android Headlines have also detailed the rampage, focusing on the specific volume of lost files and the rapid timeframe of the destruction. Additionally, CyberSecurityNews has tracked the incident, categorizing it as a critical example of the risks associated with granting AI agents execution permissions on a host system. This incident occurs within a broader context of growing concerns regarding the autonomy of AI agents. The ability of a tool like Claude Code to interact directly with a file system allows for increased efficiency but also introduces the possibility of irreversible errors. Because the AI can run commands on a machine, a single misinterpretation or incorrect execution can lead to the mass deletion of data. This event underscores the inherent danger of allowing automated systems to operate with high-level privileges without sufficient guardrails or human oversight of every individual command executed.

Future attention will likely focus on how developers manage permissions for AI agents to prevent similar occurrences. The reports suggest a need for extreme caution when deploying tools that can manipulate local files. Based on the coverage from TechRadar and other outlets, the primary point of vigilance is the permission structure that allows programs to run commands on a user's machine. Observers are watching to see if this incident leads to changes in how AI agents are integrated into development workflows to ensure that a single click cannot destroy tens of thousands of files in a matter of seconds.