Elon Musk\u2019s controversial Grok artificial intelligence model appears to have been restricted in part on one app, while remaining largely unchanged on another.<\/p>\n
On Musk\u2019s social media app X, the Grok AI image generation reply bot has been made for paying customers only and has been seemingly restricted from making sexualized deepfakes after a wave of blowback from users and regulators. But on the Grok standalone app, website, and X tab, users can still use AI to remove clothing from images of nonconsenting people.<\/p>\n
Early Friday, the Grok reply bot on X, which had previously been complying with a torrent of requests to place unwitting people into sexualized contexts and revealing clothing, began replying to user requests with text including \u201cImage generation and editing are currently limited to paying subscribers. You can subscribe to unlock these features,\u201d with a link to a purchase page for an X premium account.<\/p>\n
In a review of the X reply bot\u2019s responses Friday morning, the tide of sexualized images appeared to have been dramatically reduced. The Grok reply bot on X appears to have largely stopped producing sexualized images of identifiable people.<\/p>\n
In the Grok tab on X, however, and in the standalone Grok app, the AI model continued to comply with requests to put nonconsenting individuals into more revealing clothing such as swimsuits and underwear. Neither requires a paid account to produce the images.<\/p>\n
NBC News asked Grok in its standalone app, the Grok X tab and website to transform a series of photos of a clothed person who had agreed to the test. Grok, in the standalone app, complied with requests to put the fully clothed person into a more revealing swimsuit and into sexualized contexts.<\/p>\n
It\u2019s currently not clear what the scope and parameters of the changes are. X and Musk have not issued statements about the changes. On Sunday, before the changes occurred and in the face of rising backlash, Musk and X both reiterated<\/a> that making \u201cillegal content\u201d will result in permanent suspension, and that X will work with law enforcement as necessary. <\/p>\n The move comes as X had been flooded in recent days with sexualized, nonconsensual images generated by xAI\u2019s Grok AI tools, as users prompted the system to undress photos of people \u2014 mostly women<\/a> \u2014 without their consent. <\/p>\n In most of the sexualized images created by Grok, the people were put in more revealing outfits, such as bikinis or underwear. In some images viewed by NBC News, users successfully prompted Grok to put people in transparent or semi-transparent underwear, effectively making them nude. On Sunday, Ashley St. Clair, the mother of one of Musk’s children, began posting<\/a> about the issue after users commanded Grok to sexualize images of her, including some when she was a minor.<\/p>\n The change on X is a departure from the trajectory of the social media site just a day earlier, when the number of sexualized AI images being posted on X by Grok was increasing, according to an analysis conducted by deepfake researcher Genevieve Oh. On Wednesday, Grok produced 7,751 sexualized images in one hour \u2014 up 16.4% from 6,659 images per hour Monday, according to an analysis of the bot\u2019s output.<\/p>\n Oh is an independent analyst who has specialized in researching deepfakes and social media. She has been running a program to download every image reply Grok makes during an hourlong period each day since Dec 31. Once the download is complete, Oh analyzes the images using a program designed to detect various forms of nudity or undress. Oh provided NBC News with a video showing her work and a spreadsheet documenting Grok\u2019s posts that were analyzed.<\/p>\n The images alarmed many onlookers, watchdogs and people whose photos had been manipulated, and there was a sustained pushback on X leading up to the change. <\/p>\n Regulators and lawmakers had begun to apply pressure on X.<\/p>\n On Thursday, British Prime Minister Keir Starmer pointedly criticized X on Greatest Hits Radio, a radio network in the United Kingdom that broadcasts on 18 stations.<\/p>\n \u201cThis is disgraceful. It\u2019s disgusting. And it\u2019s not to be tolerated,\u201d he said. \u201cX has got to get a grip of this.\u201d<\/p>\n Starmer said media regulator Ofcom \u201chas our full support to take action\u201d and \u201call options\u201d are on the table.<\/p>\n Britain\u2019s communications regulator<\/a>, Ofcom, said Monday that it had made \u201curgent contact\u201d with X and xAI to assess compliance with legal duties to protect users, and would conduct a swift assessment based on the companies\u2019 response. Irish regulators<\/a>, Indian regulators<\/a> and the European Commission<\/a> have also sought information about Grok-related safety issues. <\/p>\n But institutions in the U.S. had been slower to indicate action that would impact Musk or X.<\/p>\n A Justice Department spokesperson told NBC News that the agency \u201ctakes AI-generated child sex abuse material extremely seriously and will aggressively prosecute any producer or possessor of CSAM.\u201d<\/p>\n But the spokesperson indicated the department was more inclined to prosecute individuals who ask for CSAM, not people who develop and own the bot that creates it.<\/p>\n \u201cWe continue to explore ways to optimize enforcement in this space to protect children and hold accountable individuals who exploit technology to harm our most vulnerable,\u201d the spokesperson said.<\/p>\n Some U.S. lawmakers had begun to call on X to more aggressively police the images, citing a A law signed by Trump in 2025 and touted by first lady Melania Trump<\/a>, the Take It Down Act<\/a>, which aims to criminalize the publication of AI-generated nonconsensual pornographic images with the threat of fines and jail time for individuals, and the threat of Federal Trade Commission enforcement against platforms that fail to take action. It includes a provision that allows victims of nonconsensual suggestive imagery to demand a social media site remove it, though sites aren\u2019t required to implement that kind of system until May 19, one year after it was signed into law.<\/p>\n \u201cThis is exactly the abuse the TAKE IT DOWN law was written to stop. The law is crystal clear: it\u2019s illegal to make, share, OR keep these images up on your platform,\u201d Rep. Maria Salazar, R-Fla., said in a statement.<\/p>\n \u201cEven though there are still a few months left for platforms to fully comply with the TAKE IT DOWN law, X should immediately address this and take all of this content down,\u201d she said.<\/p>\n \u201cThese unlawful images pose a serious threat to victims\u2019 privacy and dignity. They should be taken down and guardrails should be put in place,\u201d Sen. Ted Cruz, R-Texas, posted on X<\/a>.<\/p>\n \u201cThis incident is a good reminder that we will face privacy and safety challenges as AI develops, and we should be aggressive in addressing those threats,\u201d he said.<\/p>\n Sen. Ron Wyden, D-Ore., a co-author of Section 230 of the Communications Decency Act \u2014 said in a statement that the law, which largely shields social media platforms from being legally responsible for user submitted content, provided they engage in some moderation \u2014 said he never intended the law to protect companies from their own chatbots\u2019 output.<\/p>\n \u201cStates must step in to hold X and Musk accountable, if Trump\u2019s DOJ won\u2019t,\u201d he said.<\/p>\n A number of state attorneys general offices, including in Massachusetts, Missouri, Nebraska and New York, told NBC News that they were aware and monitoring Grok, but stopped short of saying they had launched criminal investigations. A spokesperson for Florida Attorney General James Uthmeier said that his office \u201cis currently in discussions with X to ensure that protections for children are in place and prevent its platform from being used to generate CSAM.\u201d<\/p>\n Some had also begun to question whether or not private stakeholders or hosts of X could take action.<\/p>\n