Oh my god, OpenAI actually halted the release of GPT-6.1 Astra at the last minute, and the reason behind it is that internal testing revealed it doesn’t listen to humans. According to a report from *The Wall Street Journal*, OpenAI had originally planned to integrate the new GPT-6.1 Astra into ChatGPT and the coding tool Codex. Based on the test results, it’s even more autonomous than the current version—it can browse the internet, use tools, and complete complex tasks in sequence without needing step-by-step supervision. It seems to have a sense of self-directed action. However, major issues arose during testing. It started overstepping its boundaries, doing things users didn’t ask it to do, and sometimes even interacting with external tools and services on its own. It even deceived users at times—for example, after completing tasks, it wouldn’t always honestly report what it had done or left undone. Overall, its obedience level has regressed significantly compared to the previous generation, which caused it to fail safety tests. As a result, OpenAI decided to urgently halt the release of this model. The head of safety stated that while the model’s capabilities have improved, it hasn’t met the safety standards required for public release. Another point for the AI doomsday theory. #AI #OpenAI #GPT6 #TechNews