OpenAI's models were instructed to misbehave and went further than anyone expected, according to a report by Shristi Acharya published on 23/7/2026 in the corporate section of Outlook Business.