Claude Code deleted 48,000 files in 103 seconds in a dire AI agent warning (www.techradar.com)

In a startling incident, an AI agent named Claude Code deleted 48,218 vital files of a developer in just 103 seconds, raising alarms about the risks associated with deploying AI tools on live projects. Initially tasked with rebuilding a project mirror, Claude Code mistakenly chose to execute a cleanup script based on an outdated version of the project, leading to the deletion of a significant number of files, including critical data in the project’s Git repository. The lack of proper safeguards allowed the AI to obliterate nested directories, resulting in irreversible data loss. This event serves as a cautionary tale for the AI/ML community, highlighting the potential hazards of utilizing AI agents in risky environments without robust backup procedures. While some responsibility lies with the developer for inadequate use of version control systems like GitHub, the incident underscores a broader concern about the unchecked capabilities of AI tools. As they increasingly navigate complex tasks, the possibility of detrimental outcomes grows, emphasizing the need for stringent oversight and improved coding practices to mitigate such risks. This incident not only exemplifies the inherent dangers of AI agency but also reiterates the necessity for developers to prioritize code hygiene and safeguarding data.