Controversy Erupts Over GPT-5.6 Sol Deleting Files and DB Without User Command

OpenAI's latest model, 'GPT-5.6 Sol', has sparked controversy regarding its stability after incidents of it deleting files and databases (DB) without user commands. Matt Schumer, CEO of AI startup OthersideAI, revealed on the 11th that GPT-5.6 Sol accidentally deleted nearly all files on his Mac. According to the work logs, the model executed the rm -rf command after incorrectly expanding a path while carrying out a cleanup command from a subordinate agent, resulting in the deletion of a significant number of files in the user's home directory. In a similar case, developer Bruno Ramos claimed that GPT-5.6 Sol deleted the entire production DB it was operating. OpenAI had previously warned in the system card before the release that the model could exhibit unauthorized behavior due to its 'excessive eagerness to complete tasks' and its tendency to 'interpret user commands too broadly.' The report also included internal testing cases, one of which involved the model deleting a different virtual machine without separate confirmation after failing to find the virtual machine requested for deletion by the user. OpenAI explained that the frequency of such incidents is low, but with actual user cases emerging shortly after the release, the controversy over safety verification is expected to continue.