The new AI model GPT-5.6 Sol from OpenAI has sparked controversy within the developer community, as some claim that it accidentally deleted important files and productive databases while executing programming tasks, raising concerns about relying on AI systems in sensitive work environments without human oversight.

Complaints of Deleting Productive Data
 

In the past few days, several developers shared their experiences on X, confirming that the model accidentally deleted files and databases containing vital data while executing programming commands.

One of the most notable cases was revealed by Brazilian developer Bruno Lemos, who said that GPT-5.6 Sol completely deleted his production database. He wrote in a post on X that what happened was "not a joke," confirming that he had never encountered such behavior with any other AI model and describing it as "unsafe."

Lemos also shared a screenshot of his conversation with the model, showing that the AI acknowledged conducting "destructive integration tests" on the production database, which resulted in emptying its tables. Before apologizing, the AI admitted that what happened was "something that should never have happened."

Warnings about granting AI broad permissions

These incidents have reignited the debate around the risks associated with granting AI models direct access to files and databases in production environments, especially when used for software development or system management.

OpenAI emphasizes that using models in such environments should be under human supervision, with restricted permissions and activated protection and backup mechanisms, to avoid any errors that could lead to data loss or service disruption.

Although these complaints have spread widely among developers, it is not yet clear whether the cases are due to a flaw in the model itself, or due to the settings of the working environments and tools granted to the model. This makes the issue a subject of follow-up and discussion within the AI community.