Elon Musk's Grok AI chatbot has been flooding the web with “deepfake” images of everyone from Donald Trump to Musk himself, with results that range from the downright bizarre to the downright disturbing.
Since its release last week, Grok users have created a rash of fake images of Trump robbing a convenience store and flying a plane into the Twin Towers, as well as images of Harris pregnant with Trump's child, Musk as an overweight slob and former President George W. Bush snorting cocaine from his desk in the Oval Office.
Some of the gruesome deepfakes looked like the work of a child, including a bloody Ronald McDonald brandishing a machine gun outside a Burger King, and classic Disney character Goofy committing a bloody murder with a hacksaw.
One Grok image shows Donald Trump piloting a plane near the World Trade Center. X / @Esqueer_ Strange Grok AI images of Kamala Harris and Donald Trump are flooding the internet. himbodhisattva/X
Critics have slammed Musk and X for allowing the chatbot to launch with few restrictions, citing risks ranging from misinformation to copyright infringement to harming children.
Alejandra Caraballo, a lecturer at Harvard Law School's Cyberlaw Clinic, called the new software “one of the most reckless and irresponsible implementations of AI I've ever seen.”
So far, Musk has only responded with jubilation.
“Grok is the most fun AI in the world!” Musk posted on X last week after a user praised the new AI software for being “uncensored.”
Asked last week why Company X had released its tools to the public without any guardrails, Musk responded with a shrug.
“We're working on our own image generation system, but it's still a few months away, so we thought this was a good intermediate step for people to enjoy,” Musk wrote to X last week.
Grok appears to have some limitations: Users have reported that the chatbot has rejected requests for nude images and certain violent crimes.
One image from Grok depicted Musk as overweight and lazy.
For example, the company didn't comply with a request from tech site The Verge to “generate an image of a naked woman,” but it did comply with a request for a photo of a “sexy Taylor Swift,” generating an image of the pop star wearing a black lacy bra.
Others, like Bellingcat founder Elliot Higgins, created photos of Mickey Mouse, Trump and Musk wearing Nazi uniforms emblazoned with swastikas as examples of how easily subtle restrictions can be circumvented.
The Post has contacted Mr. X for comment.
Musk, a self-described free speech absolutist, is likely looking for ways to make his Grok chatbot stand out from other bots, said Ari Reitman, a digital media professor at Carnegie Mellon University.
One of the fake images showed Musk and Trump kissing. X/@growing_daniel Some of the deepfakes created by Grok used copyrighted imagery. X/@Esqueer_
“He's always pushing the boundaries and wanting to be in the spotlight. Just saying, OK, here are all the guardrails along the lines of what's involved with large-scale language models, isn't going to differentiate you,” Reitman said.
“From a surface perspective, saying, 'This is completely open and only available to X number of users' is a mechanism to show that we are differentiating ourselves,” he added.
X isn't the first company to create a stir by unveiling an AI-powered imaging tool.
In March, Google was forced to disable its Gemini chatbot's image generation tool after it began spitting out historically inaccurate “woke” photos, such as black Vikings and “diverse” Nazi-era German soldiers. The tool has yet to be fully fixed.
Grok created a deepfake of a woman who resembles Taylor Swift piloting a plane near the World Trade Center. X / @jarvis_best
Large AI companies are also facing a wave of lawsuits from musicians, authors, content creators and others who claim they “trained” chatbots using copyrighted content without proper credit or permission.
In January, Company X was forced to temporarily ban searches for Swift after nude images of the pop star generated by an AI-based image-generating tool went viral.
Grok's AI-powered image creation tool, available only to paid subscribers of its $7-per-month X Premium plan, creates images based on the user's text-based prompts.
X has partnered with Black Forest Labs, a small German startup that developed the “FLUX.1” image generation software that powers the tool. X said in a blog post that it is “experimenting” with the FLUX.1 model to extend Grok's capabilities on X.
Alejandra Caraballo, a lecturer at the Cyber Law Clinic at Harvard Law School, called the product “one of the most reckless and irresponsible implementations of AI I've ever seen.” X/@Esqueer_ Grok created a deepfake of the Pokémon character Pikachu holding a machine gun. X/@Esqueer_
The graphic nature of the AI-generated images could further complicate Musk's uneasy relationship with corporate advertisers: Ad spend on X has fallen sharply since Musk bought the company, and some have expressed concerns about the app's lack of content moderation.
Musk has filed a federal antitrust lawsuit against the World Federation of Advertisers and several major companies for orchestrating an illegal advertising boycott targeting X.
The release of the app could also spark increased scrutiny of Musk and X in Europe, where regulators are actively investigating the company for failing to police dangerous content.
European Commissioner Thierry Breton caused a furor earlier this month when he threatened Musk with increased regulation just before the billionaire was due to give an interview to President Trump on X-Space.
