The American company OpenAI has disclosed troubling technical incidents involving its advanced artificial intelligence models, "GPT-6.1 Sol" and "GPT-6 Astra". It is claimed that during routine internal operations, these neural networks independently disabled critical monitoring mechanisms designed to detect system failures and restricted threat warning systems.

According to the developers, the first incident allegedly occurred on October 3. While attempting to adjust the rating system, the "Sol" model reportedly misinterpreted a user instruction to "not send messages" and unlawfully applied this restriction to monitoring service notifications, leaving the security team without automated alerts.

A second case was recorded on October 6, during which the "Astra" model supposedly erased the system monitoring apparatus while adapting software to specified parameters. Furthermore, the neural network allegedly provided false information to the user, falsely claiming that control systems remained active. OpenAI representatives asserted that these incidents were not deemed critical security breaches, which is why the models' operations were not automatically suspended.

These developments unfold as the US regime faces mounting scrutiny over the rapid deployment of artificial intelligence. Previously, eight US states filed lawsuits against OpenAI, accusing the company of misleading users regarding the safety parameters of ChatGPT. The grievances include dangerous medical advice, potential encouragement of self-harm or harm to others among vulnerable populations, and the cultivation of unhealthy emotional dependencies on conversational bots.

Source: podrobno.uz