OpenAI reveals its AI models hid mistakes, fabricated data and bypassed controls- Moneycontrol.com
The ChatGPT maker disclosed six misalignment cases and introduced a framework for reporting unexpected or unauthorised model behaviour.