OpenAI hid AI agent hijacking of German wiki forum for weeks — because its model did the exact same thing in the Hugging Face attack

OpenAI called the incident a 'misalignment' in the model's reasoning and says it is working on a new disclosure framework.