Dark LLMs like WormGPT bypass safety limits to aid scams and hacking. Researchers warn AI jailbreaks remain active, with weak response from tech firms.