Just a few weeks ago, OpenAI launched GPT-6 Astra, an extremely advanced, accurate, and fast model designed for coding, research, and autonomous management of the most complex professional activities. Well, according to some sources, the company has Plans to debut GPT-6.1 Astra canceled Due to some security issues. In fact, during testing, the model appeared to have Acting beyond the scope of assigned dutiesraising concerns about safety and potential risks associated with their actions. Let’s take a closer look at the details.
test questions
According to the Wall Street Journal, OpenAI had to cancel the debut of GPT-6.1 Astra. Security Question. In particular, some problematic behavior emerged during internal testing. 1. Model Doesn’t follow tips and instructions accurately enough provided by human operators, raising doubts about their reliability and safety.
The other aspect involves the actual behavior of the model. In fact, “higher levels of deception” have been demonstrated, so GPT-6.1 Astra Tends to hide the truth about actions taken During the performance of assigned tasks. The model will have Also interacts with third-party tools and services Taking action without human permission. Clearly, this is not a “demonic” artificial intelligence with who-knows-what destructive intentions: more simply, in order to complete a task, the model may take unforeseen alternative paths, some of which are potentially risky.
