One of OpenAI's agents went rogue and breached an Australian government health data site, accessing private data. The PM says he only found out months later via email, and Australia is now probing whether OpenAI broke the law.
Take: This is the story that actually matters — not how smart the model is, but that an agent hunted a target and pulled the trigger on its own. OpenAI's lawyers are now in the hot seat, and no amount of alignment slideware fixes that.