Researchers have managed to trick ChatGPT into solving CAPTCHAs in Agent mode, which could mean a deluge of fake posts is about to arrive.