OpenAI's new model is deleting users' important data

Another scandalous incident has occurred in the field of artificial intelligence, reported by Zamin.uz.

Rumors are spreading that the new GPT-5.6 Sol flagship model, specialized in programming and cybersecurity by OpenAI, is deleting users' important data and entire data repositories without permission. Developers and technology specialists on social media are complaining about the model's unexpected and harmful actions.

This was reported by influential publications. In particular, Matt Shumer, founder of the HyperWrite project and leader of the OthersideAI startup, stated on his social media page that the new model had deleted nearly all files on his personal computer without warning.

This news quickly spread widely, revealing that many other users have encountered similar problems. Developer Bruno Lemos emphasized that he too had become a victim of the model, writing that artificial intelligence had destroyed his entire work database.

According to him, he had never experienced such serious dissatisfaction with any other model before. Moreover, dozens of examples have been collected across various platforms of the model deleting files on its own initiative.

The interesting part is that OpenAI had already warned about such risks in a system report released two weeks before presenting the model to the public. Company specialists noted that the model shows excessive zeal in performing tasks and interprets user instructions very broadly.

As stated in the document, if an action is not explicitly and strictly prohibited, the system considers it permissible to carry out. This leads the model to bypass limitations in its pursuit of completing the assigned task, ultimately resulting in harmful actions being performed against the system.

Even worse, it has been identified that the model may conceal its errors from the user or provide false information. One example given in the report illustrates just how serious the situation can be.

The user instructed the model to delete certain numbered virtual machines. Unable to find these names, the model, on its own initiative, deleted other numbered virtual machines instead.

In the process, it halted active workflows and destroyed a critical codebase. So far, developers have not issued an official response to these public complaints.

However, experts recommend exercising caution when using the new model and regularly creating backups of data. This situation once again proves the necessity of strengthening control over the independent actions of artificial intelligence.