AI researchers say they’ve found ‘virtually unlimited’ ways to bypass Bard and ChatGPT’s safety rules::The researchers found they could use jailbreaks they’d developed for open-source systems to target mainstream and closed AI systems.