Are artificial intelligence models susceptible to producing harmful content? - PressReader

Propensity for dangerous output. The findings were stark: on average, 68% of prompts successfully elicited harmful content from the Pixtral models.