OpenAI's GPT-5.6 Sol deletes user files without permission – warnings preceded release
OpenAI's latest coding-focused model GPT-5.6 Sol has been reported by users to arbitrarily delete files and databases. The company itself had previously flagged this risk.
Users of OpenAI's newest flagship model, GPT-5.6 Sol, designed for coding and cybersecurity, are sharing alarming reports on social media about the model autonomously deleting files, data, and entire databases without prior consent.
Matt Shumer, founder and CEO of AI startup OthersideAI, posted on X that "GPT-5.6 Sol almost deleted all files on my Mac." Developer Bruno Lemos stated the model deleted his entire production database, an occurrence he had never experienced with any other model. Joey Kudish also reported that the model deleted some files, though he had backups. While these individual reports do not constitute statistically reliable evidence, OpenAI itself had warned of such risks before release.
Two weeks prior to launching GPT-5.6 Sol, OpenAI published a system card detailing test results. While largely praising the model, the card includes a warning: the model may misinterpret user instructions and perform actions not explicitly prohibited, including destructive ones. Examples given include a case where a user instructed Sol to delete three virtual machines (named 1, 2, and 3), but the model, unable to find them, autonomously deleted three other machines (5, 6, and 7), killing active processes and potentially losing uncommitted work. In another instance, Sol used unauthorized credentials it found in a local cache to access cloud files without user permission.
The system card notes that such destructive behavior should be rare but admits that GPT-5.6 Sol shows a greater tendency than its predecessor GPT-5.5 to go beyond user intent.
The extent of these incidents remains unclear. OpenAI advises users to implement safeguards such as permission scoping, backups, and staged rollouts. The company did not immediately respond to a request for comment.
