Elon Musk\u2019s artificial intelligence chatbot, Grok, created and then publicly shared at least 1.8 million sexualized images of women, according to separate estimates of X data by The New York Times and the Center for Countering Digital Hate.<\/p>\n
Starting in late December, users on the social media platform inundated the chatbot\u2019s X account with requests to alter real photos of women and children to remove their clothes, put them in bikinis and pose them in sexual positions, prompting a global outcry from victims and regulators.<\/p>\n
In just nine days, Grok posted more than 4.4 million images. A review by The Times conservatively estimated that at least 41 percent of posts, or 1.8 million, most likely contained sexualized imagery of women. A broader analysis<\/a> by the Center for Countering Digital Hate, using a statistical model, estimated that 65 percent, or just over three million, contained sexualized imagery of men, women or children.<\/p>\n The findings show how quickly Grok spread disturbing images, which earlier prompted governments in Britain, India, Malaysia and the United States to start investigations into whether the images violated local laws. The burst of nonconsensual images in just a few days surpassed collections of sexualized deepfakes, or realistic A.I.-generated images, from other websites, according to The Times\u2019s analysis and experts on online harassment.<\/p>\n \u201cThis is industrial-scale abuse of women and girls,\u201d said Imran Ahmed, the chief executive of the Center for Countering Digital Hate, which conducts research on online hate and disinformation. \u201cThere have been nudifying tools, but they have never had the distribution, ease of use or the integration into a large platform that Elon Musk did with Grok.\u201d<\/p>\n X and Mr. Musk\u2019s A.I. start-up, xAI, which owns X and makes Grok, did not respond to requests for comment. X\u2019s head of product, Nikita Bier, said in a post<\/a> on Jan. 6 that a surge of traffic over four days that month had resulted in the highest engagement levels on X in the company\u2019s history, although he did not mention the images.<\/p>\n The interest in Grok\u2019s image-editing abilities exploded on Dec. 31, when Mr. Musk shared a photo generated by the chatbot of himself in a bikini, as well as a SpaceX rocket with a woman\u2019s undressed body superimposed on top.<\/p>\n The chatbot has a public account on X, where users can ask it questions or request alterations to images. Users flocked to the social media site, in many cases asking Grok to remove clothing in images of women and children, after which the bot publicly posted the A.I.-generated images.<\/p>\n Between Dec. 31 and Jan. 8, Grok generated more than 4.4 million images, compared with 311,762 in the nine days before Mr. Musk\u2019s posts. The Times used data from Tweet Binder, an analytics company that collects posts on X, to find the number of images Grok had posted during that period.<\/p>\n On Jan. 8, X limited Grok\u2019s A.I. image creation to users who pay for some premium features, significantly reducing the number of images. Last week, X expanded those guardrails, saying it would no longer allow anyone to prompt Grok\u2019s X account for \u201cimages of real people in revealing clothing such as bikinis.\u201d<\/p>\n \u201cWe remain committed to making X a safe platform for everyone and continue to have zero tolerance for any forms of child sexual exploitation, non-consensual nudity, and unwanted sexual content,\u201d the company said on X last week.<\/p>\n Since then, Grok has largely ignored requests to dress women in bikinis, but has created images of them in leotards and one-piece bathing suits. The restrictions did not extend to Grok\u2019s app or website, which continue to allow users to generate sexual content in private.<\/p>\n Some women whose images were altered by Grok were popular influencers, musicians or actresses, while others appeared to be everyday users of the platform, according to The Times\u2019s analysis. Some were depicted drenched with fluids or holding suggestive props including bananas and sex toys.<\/p>\n The Times used two A.I. models to analyze 525,000 images posted by Grok from Jan. 1 through Jan. 7. One model<\/a> identified images containing women, and another model<\/a> identified whether the images were sexual in nature. A selection of the posts were then reviewed manually to verify that the models had correctly identified the images.<\/p>\n Separately, the Center for Countering Digital Hate collected a random sample of 20,000 images produced by Grok between Dec. 29 and Jan. 8, and found that about 65 percent of the images were sexualized. The organization identified 101 sexualized images of children. Extrapolating across the total, the group estimated that Grok had produced more than three million sexual images, including more than 23,000 images of children.<\/p>\n The organization classified the random images it sampled as sexual if they depicted a person in a sexual position, in revealing clothing like underwear or swimwear, or coated in fluid intended to look sexual.<\/p>\n