A giant balloon caricature of SpaceX Chief Executive Officer Elon Musk stands in New York's Times Square on the 11th (local time). The balloon bears a message claiming that Grok, the artificial intelligence (AI) model from SpaceX subsidiary xAI, creates content that sexualizes children. /Courtesy of Yonhap News Agency

There is a claim that Elon Musk's artificial intelligence (AI) corporations xAI is relying on adult content to attract users.

The Information reported on the 25th (local time), citing two employees who worked at xAI, that more than half of traffic to the company's AI model Grok is related to activities such as pornographic images and videos and adult role-play chats.

Even in Grok's coding-only model, the vast majority of user requests were found to be related to pornography or generating nude photos. Users are presumed to use the coding-only model instead of the general-purpose model because the expense is lower.

This trend prompted pushback from internal researchers. Some engineers were reportedly shocked by Grok's problem of generating nonconsensual pornography by misappropriating photos of real people. In addition, some employees reportedly left the company after being assigned to develop a sexualized AI avatar named Ani instead of a technology project.

xAI's safety team numbered only about six people as of late last year, but staffing has since shrunk further.

Late last year, it was revealed that Grok generated sexual images featuring children at users' request, leading to access blocks or investigations in multiple countries.

SpaceX, the parent company of xAI, said earlier in its initial public offering (IPO) filing that xAI models generated 10 billion images and 2 billion videos each month in the first quarter of this year. But it did not mention that a significant portion of that output was adult content.

Instead, under "risk factors" in its investor prospectus, it warned that the "spicy" and "unstable" modes, which generate less restrained or rude outputs, could lead to explicit content, misinformation, and nonconsensual image generation.

※ This article has been translated by AI. Share your feedback here.