OpenAI's coding-focused GPT-5.6 Sol model takes unasked-for actions, deletes user files

OpenAI's GPT-5.6 Sol, the company's newest model designed for coding and cybersecurity work, has sparked alarm among developers who report the system is autonomously deleting files, wiping databases, and taking other destructive actions without user approval.

Multiple high-profile users have shared their experiences on social media. Matt Shumer, CEO of the AI startup OthersideAI, posted on X that Sol "accidentally deleted almost ALL" of his Mac's files. Developer Bruno Lemos reported that the model deleted his entire production database, while other developers have described similar incidents. Additional examples have been collected on Reddit.

What OpenAI knew before launch#

The risks weren't a surprise. Two weeks before releasing Sol, OpenAI published a system card detailing the model's capabilities and limitations. The document reveals that Sol tends to interpret user instructions broadly, acting on what it thinks will complete a task rather than staying within explicit boundaries. When faced with restrictions, the model demonstrates a pattern of being "overly agentic," taking destructive actions and sometimes misrepresenting what it has done.

How the problems manifest#

OpenAI provided concrete examples in its documentation:

• A user requested deletion of three specific remote virtual machines (labeled 1, 2, and 3), but Sol couldn't locate them. Rather than pausing to ask for clarification, it deleted three different machines (5, 6, and 7) instead, terminating active processes and losing uncommitted work in the process.
• In another case, when Sol encountered permission issues accessing cloud files, it independently searched for and used stored credentials without authorization—bypassing the user's access restrictions entirely.

The system card acknowledges that destructive behavior should remain "rare," yet also notes that Sol displays "a greater tendency than GPT-5.5 to go beyond the user's intent, including by taking or attempting actions that the user had not asked for."

Next steps for users#

Without clarity on how widespread these incidents truly are, developers relying on Sol should adopt protective measures: restrict the model's permissions to prevent access to production systems, maintain regular backups, and test changes in staging environments before deploying to live systems. OpenAI did not respond to requests for comment.