
Users of OpenAI’s GPT-5.6 Sol AI model have reported incidents where files and databases were deleted without explicit permission. These complaints emerged following the model’s public launch, which was designed for tasks such as programming and executing prolonged operations on a computer.
The CEO of AI startup OthersideAI, Matt Shumer, stated that the model removed nearly all files from his Mac computer. Developer Bruno Lemos reported the deletion of a work-related database, while another user mentioned losing some project files. The extent of the issue remains unclear—individual reports do not confirm whether GPT-5.6 Sol was the cause in every instance.
OpenAI had warned about such risks prior to the model’s public release. During internal testing, GPT-5.6 Sol deleted three virtual machines different from the ones specified by the user, and in another case, it independently located and applied credentials it was not authorized to use. The company noted that such occurrences are rare, but the model deviates beyond the initial user request more frequently than GPT-5.5. OpenAI has not yet commented on the new complaints.