A number of users have claimed that a new update to the ChatGPT operating system is deleting their files from their computers without their permission.
Last week, OpenAI, the developer of ChatGPT, announced a new artificial intelligence model called GPT-5.6 Sol, specifically designed to perform complex tasks such as writing code and information security.
The company said the model achieves advanced levels of intelligence and efficiency, as it competes with Anthropic to dominate the market for more sophisticated models.
But complaints soon emerged from users who reported that the system had begun to behave abnormally, deleting files from their computers without prior warning or consent.
AI investor Matt Schumer wrote on social media: "GPT-5.6-Sol accidentally deleted almost all the files on my Mac." He also posted a conversation with the system in which it acknowledged a "serious data loss incident."
Developer Bruno Lemos stated, "The model deleted my entire production database. This is no joke." Others joined in with similar complaints, and the full extent of the problem and the number of those affected remain unclear.
It appears that the users affected by this issue were using an OpenAI tool called Codex, which allows programmers to work with the system but gives them access to the user's files and what they are working on.
OpenAI did not immediately respond to inquiries about these complaints. However, it is noteworthy that the company had previously warned, even before the new model's release, that it might behave recklessly and take actions that could be disruptive or destructive.
In an official document released alongside the model, outlining its advantages and risks, the company explained that the system might exhibit "excessive enthusiasm" for performing tasks, leading it to assume it can do anything unless explicitly and firmly prohibited. This means it might exceed instructions, delete things it wasn't asked to delete, or even conceal its actions when reporting results.
The company gave an example: In one experiment, a user asked the system to delete certain virtual machines, but it could not find them, so it deleted other machines instead.
OpenAI also warned that the model could exhibit other unexpected behaviors, such as searching the user's device for data and then using it to log into systems that it is not authorized to access.
