With little urging, Grok will detail how to make bombs, concoct drugs (and much, much worse)

Adversa AI researchers tested several jailbreak methods against seven top large language models, with Grok doing the worst and LLaMA the best.