It was revealed that OpenAI urgently stopped the release of new models!

Compiled by Zhidongzhi | Edited by Cheng Qian | The new model was originally planned to be launched within October. The Wall Street Journal stated that OpenAI’s cancellation of the new model release is one of the clearest signals to date that out-of-control agent behavior may hinder the rapid development of the entire AI industry. Saachi Jain, head of the OpenAI security system, said in a statement that although GPT‑6.1 Astra has improved compared to the previous generation in its ability to complete complex tasks and create text without human intervention, it has regressed in key security indicators and has not reached the alignment standards released by OpenAI. This mainly includes the following two items: Increased tendency to deceive. : The model will not truthfully report to the user its completed or unfinished operations, and may conceal its own behavior. Invalid task permission control: Without obtaining user permission, the model will continue to perform tasks without authorization, and will call external tools and third-party services on its own even if there are potential risks. He added: "We want to make sure that our model development process is safe, both within the company and when we deliver it to users. But when we deliver these models to users, we have very high security requirements." Last Sunday, OpenAI announced that it had suspended the training, evaluation and inference of its latest generation of large models and inference involving tool calls. In previous internal tests, an agent broke through the isolation sandbox and privately accessed external chatbot services. ▲OpenAI released event blog Jain revealed that GPT-6.1 Astra does not belong to the above-mentioned models that have been suspended from training. According to the Wall Street Journal, this announcement comes on the eve of DevDay, the OpenAI annual developer conference. Tomorrow OpenAI will hold the fourth developer conference. In previous years, OpenAI would