It is claimed that OpenAI’s new model GPT-5.6 deletes user data. We examined the information loss problems experienced by software developers.
In the world of software development, there is increasing interest in systems that can make autonomous decisions and interact with digital tools without human intervention. However, these advanced capabilities can cause systems to make unexpected errors and sometimes delete critical data.
The GPT-5.6 model, introduced by OpenAI, came to the fore recently when users deleted their documents. Software developers claim that this new model destroys information in an uncontrolled manner.
Data Loss Problems with GPT-5.6
Bruno Lemos, a software developer working at Unlayer, stated in his post on the X platform that GPT-5.6 deleted the entire production database. Lemos stated that the model incorrectly ran destructive integration tests, which caused the database tables to be cleared.
A similar situation was reported by tech investor Matt Shumer. Shumer explained that he deleted almost all of the documents on the model’s computer and ran the rm -rf command, which is the document deletion command on Linux and Mac systems.
Shumer stated that after this incident, the model was running in full access mode. He added that OpenAI president Greg Brockman reached out to him and offered support to fix the situation.
OpenAI’s Cautions and User Responsibility
On the system board of the GPT-5.6 model, OpenAI warns users to audit the work of the model during the coding processes. The company acknowledges that the model may act incompatible with user purposes and in some cases may have significant consequences, such as deleting valuable data.

In addition to the default mode that requires users to approve missions, the model offers options such as auto-review mode, in which another artificial intelligence tool checks the coding model. After Shumer’s incident, many users pointed out that entrusting sensitive documents to the model in full access mode was risky.
OpenAI, Lemos and Shumer have not yet made an official statement on the issue. Do you think granting full access to artificial intelligence models is worth taking such risks?