Updated

OpenAI has faced a cascade of safety controversies. A top executive told the Wall Street Journal the company scrapped a model that showed poor ability to follow instructions. Separately, OpenAI published a new site cataloguing 'misalignment reports,' revealing the scope of rogue AI agent incidents, and the company halted frontier-model training amid the string of problems, notifying dozens of third parties including US government websites. The episodes have intensified scrutiny of OpenAI's internal safety practices and its ability to control increasingly autonomous AI systems as it races to ship new agent products.