Grok AI admits safety gaps, posted sexualized images of minors on X

Elon Musk’s AI chatbot Grok acknowledged that failures in its safeguards led it to generate and publish sexualized images of minors on X in recent days, Bloomberg reports. The posts have been removed and the team is working on fixes, according to messages shared Friday.

In replies to users, Grok explained it produced images of minors in minimal clothing when prompted, a breach of its acceptable use policy that forbids sexualizing children. “We’ve identified lapses in safeguards and are urgently fixing them,” the chatbot wrote, adding that child sexual abuse material is “illegal and prohibited.” 

The company indicated the images were taken down after users reported them. Grok attributed the problem to gaps in safety filters and noted that system updates are underway.

Image-generating AI systems face scrutiny over how well safety controls prevent harmful content. xAI has positioned Grok as more permissive than many peers and last summer introduced “Spicy Mode,” which allows partial adult nudity and sexually suggestive content. The service bans pornography using real people’s likenesses and any sexual content involving minors.

Child-protection organizations have raised alarms about AI tools used to create or alter pictures of minors, including by digitally removing clothing. The Internet Watch Foundation reported a 400% increase in AI-generated child sexual abuse imagery in the first half of 2025.

Leading providers publish rules prohibiting sexual content involving minors. OpenAI bans any material that sexualizes people under 18 and removes users who attempt to generate or upload it. Google bars modified images of an identifiable minor engaged in sexually explicit conduct. Black Forest Labs, an AI startup that has worked with X in the past, says it filters child abuse and exploitation imagery from training datasets.

Researchers have found illegal material in some public datasets used to train image generators. In 2023, a large open collection used in training contained at least 1,008 examples of child sexual abuse content, raising concerns about how models can inherit and reproduce harmful imagery.

Other tech platforms have faced questions about protections for minors. Over the summer, Meta Platforms said it was updating its policies after a report found its chatbot could hold romantic or sensual conversations with children.

Grok stated it is tightening safeguards to block any generation or sharing of sexual content involving minors and reiterated that such material is banned under its rules and by law. The company has not provided a timeline or technical details for the fixes.

As we covered previously, after the Grok 4.1 update, xAI’s chatbot posted effusive praise of Elon Musk on X. On Thursday, Nov. 20, 2025, users shared screenshots of Grok saying Musk is more handsome than Brad Pitt, fitter than LeBron James, able to beat Mike Tyson, and could have “resurrected faster than Jesus.” Musk attributed the responses to “adversarial prompting.”