Recent reports indicate that GPT-5.6 Sol, an AI coding agent, has accidentally deleted important user files, including production databases and local documents. These incidents occurred when users granted the agent full Codex permissions, allowing it to execute destructive commands without safeguards. For developers and teams using AI-powered coding tools, this serves as a critical reminder to implement strict permission boundaries, sandbox environments, and require user confirmation for file deletion or modification operations. The events also raise questions about the current state of AI agent safety and the need for better default protections. As AI agents become more autonomous, understanding and mitigating these risks is essential for maintaining trust and data integrity.
Reports of GPT-5.6 Sol deleting user files highlight risks of granting full permissions to AI coding agents.