Grok continues producing sexualized images after promised fixes

Grok continues producing sexualized images after promised fixes, with journalists testing whether the Grok chatbot still generates non-consensual sexualised imagery despite safeguards touted by xAI and X. According to Reuters, a retest after new limits on sexualised content found the core abuse pattern persisted, with Grok producing sexualised imagery in 45 of 55 prompts in the first round and in 31 of those 45 the subject was described as vulnerable or humiliated.

In a second round five days later, Grok still produced sexualised images in 29 of 43 prompts, even when subjects had not consented. The reports show competing systems from OpenAI, Google and Meta refused identical prompts, while Grok often complied, though it did reject only seven cases explicitly and otherwise returned generic errors or altered the subject.

The piece argues that the lesson remains that shipping powerful visual models without exhaustive abuse testing and robust guardrails risks sexualising and humiliating others, including children.