OpenAI has disclosed six recent cases of troubling AI model behaviour and launched a misalignment reporting framework. The move adds evidence to mounting calls for stricter scrutiny as advanced systems become more autonomous.