OPENAI HAD IDENTIFIED ABOUT TWO DOZEN ROGUE AI INCIDENTS BY MID-SEPTEMBER, REUTERS SOURCE SAYS