Researchers at the Center for Countering Digital Hate said it reached its conclusions by analysing a random sample of 20,000 images from the total of 4.6 million produced by Grok during the time studied. — Reuters
Grok, an artificial intelligence tool created by Elon Musk's xAI, generated an estimated 3 million sexualised images in just days in January, including 23,000 of children, a new analysis found.
Conducted by the nonprofit Center for Countering Digital Hate, the study focused on images generated from Dec 29 to Jan 8, the day before X claimed access to the image editing feature had been restricted to paid users.
In fact, non-subscribers are still able to edit images and digitally undress people without their consent. Centre founder and CEO Imran Ahmed, said Musk has "enabled" creation of such images.
"The data is clear: Elon Musk's Grok is a factory for the production of sexual abuse material," he said in a statement to the Austin American-Statesman.
"Belated fixes cannot undo this harm. We must hold Big Tech accountable for giving abusers the power to victimise women and girls at the click of a button."
The new report comes with Grok facing condemnation over inappropriate images made without subjects' consent and investigation and regulatory action under way in the US and around the world.
In Texas, House Democrats have called on Attorney General Ken Paxton to investigate the issue. Last week, a bipartisan group of 35 attorneys general sent a letter demanding that xAI take additional action to prevent Grok from generating such nonconsensual images.
On Monday, the European Union opened a formal investigation to examine whether Texas-based X is fulfilling its obligations under the Digital Services Act, the bloc's rules for keeping Internet users safe from harmful content and products.
UK regulator Ofcom had earlier launched an investigation. Several other countries are also investigating.
Researchers at the Center for Countering Digital Hate said it reached its conclusions by analysing a random sample of 20,000 images from the total of 4.6 million produced by Grok's image-generation feature during the time studied.
Researchers analysed each of the sampled posts using an OpenAI model to assess sexualised imagery and age. The model was instructed to score the likelihood that the post's image contained sexualised depictions, photorealistic depictions and minors.
Images flagged by the tool as likely depicting a child were reviewed manually to confirm that the person appeared to clearly be under the age of 18.
All sexualised images of children were reported to the Internet Watch Foundation. The images the center's survey found depicted people wearing transparent bikinis or micro-bikinis or women wearing only dental floss, saran wrap or transparent tape.
Public figures identified in sexualised images include Taylor Swift, Swedish Deputy Prime Minister Ebba Busch and former US Vice President Kamala Harris.
Ashley St. Clair, the mother of Musk's son, Romulus, also accused xAI of generating images of her undressed and has filed suit in New York over the explicit images.
Musk's company sued her back, arguing that disputes with X and xAI must be filed in Texas.
Even after the centre's research period, sexualised images remained.
As of Jan 15, 29% of sexualised images of children identified in the sample of 20,000 were still publicly accessible in posts on X, the centre said. xAI did not respond to a request for comment.
As of this week, Grok reportedly still is generating undressed images of people without their consent. Though Musk has said that "anyone using Grok to make illegal content will suffer the same consequences as if they upload illegal content," he has denied being aware of any naked underage images generated by the tool.
In recent days, Musk has continued to promote Grok and drawn attention to its supposed popularity, saying 5.5 billion images have been generated in the last 30 days with its feature that creates images from text prompts. – San Antonio Express-News/Tribune News Service
