Grok AI was used to generate around 3 million sexualised images over an eleven day period earlier this year, including of children.
Researchers from the U.S. Center for Countering Digital Hate (CCDH) announced last week the tool created child sexual abuse material every 41 seconds during this period.
The findings come after Elon Musk faced global pressure to restrict Grok’s capabilities on X, the social media platform he owns.
Here’s what you need to know.
Context
Grok is an AI tool developed by Musk’s company xAI for use on X.
Late last year, users discovered they could request Grok edit images of people with prompts like “put her in a bikini”.
Musk himself participated initially, reposting generated images of himself and others, including Bill Gates, in bikinis.
Study
CCDH researchers estimate Grok created 3 million sexualised images of people between 29 December and 8 January.
Images were classified as sexualised if they showed people in sexual positions, revealing clothing such as underwear or swimmers, or depicted bodily fluids.
The researchers found sexualised images depicting public figures including musicians, actors, and politicians.
Your contribution ensures The Daily Aus can continue doing the work you love.
I want to give $10 monthly to help TDA keep doing the news.
Your details:
Grok was also used to create child sexual abuse material: an average of one image every 41 seconds over the 11 days.
Global response
Following widespread reports of sexual abuse material on X, Indonesia and Malaysia became the first countries to block Grok earlier this month.
UK communications regulator Ofcom has launched a formal investigation, with potential fines of up to £18 million ($AU35 million) or 10% of the company’s global revenue.
Australia’s eSafety Commissioner also announced there had been an “increase from almost none to several reports” of non-consensual sexual images being created on the site.
X response
In early January, the official account for Grok posted admitting to “lapses in safeguards.”
Musk also commented, saying: “Anyone using Grok to make illegal content will suffer the same consequences as if they upload illegal content”.
X later announced it had stopped Grok from editing people’s pictures to show them in “revealing clothing”.
However, research lab AI Forensics has found users are still able to create sexualised images by accessing Grok directly through its website, rather than through X.
In a post to X on Monday, the official Grok account announced users had generated more than 5.5 billion images in the last 30 days.
