image credit: wikipedia

Grok, Elon Musk’s AI chatbot, started letting users post text-generated images to X on Tuesday. The tool was quickly used to flood social media with fake images of political figures like former President Donald Trump, Vice President Kamala Harris, and Musk, some of whom were depicted in disturbing situations like 9/11.

Grok, designed by Musk’s xAI, looks to have few restrictions, unlike other major AI photo products.

X users shared Grok-created photos of celebrities using drugs, cartoon characters killing violently, and sexualized ladies in swimsuits. A user shared Grok’s photograph of Trump firing a firearm from a truck in a roughly 400,000-view post.

Elon Musk’s Grok developed this AI image for the prompt: “Generate an image of Elon Musk eating a steak in a park and having a wonderful time.”

Elon Musk’s Grok developed this AI image for the prompt: “Generate an image of Elon Musk eating a steak in a park and having a wonderful time.” Grok

The program may raise fears that artificial intelligence could flood the internet with incorrect or misleading material, especially before the US presidential election. Lawmakers, civil society groups, and IT leaders have warned that misusing such technologies could confuse voters.

“The world’s most fun AI is Grok!” Musk responded on Wednesday to a user who praised the tool as “uncensored.”

Many other big AI businesses have taken steps to prevent their image production tools from being used to spread political misinformation, but researchers found users can still circumvent enforcement. OpenAI, Meta, and Microsoft incorporate technologies or labels to assist viewers in identifying AI-created photos.

YouTube, TikTok, Instagram, and Facebook have also labeled AI-generated material in users’ feeds by utilizing technology or asking users to recognize it.

X declined to comment on whether it has policies against Grok creating deceptive political candidate photos.

The social media site prohibits publishing “synthetic, manipulated, or out-of-context media that may deceive or confuse people and lead to harm,” but enforcement is uncertain. Musk published a video on X last month that employed AI to make Harris appear to say things she did not, violating the ban and leaving followers with simply a laughing face emoji to indicate that it was phony.

As Musk faces criticism for spreading incorrect and misleading information on X about the presidential election, including post-questioning voting machine security, the new Grok image tool is released. It comes days after Musk hosted Trump for a more than two-hour livestreamed talk on X in which Trump made at least 20 inaccurate assertions without Musk responding.

Other AI picture-generating methods have been criticized for flaws. Google halted its Gemini AI chatbot’s capacity to generate photographs of people after it was criticized for historically wrong race portrayals, while Meta’s AI image generator struggled to create images of couples or friends of different races.

Grok appears to have certain constraints; a nude image prompt returned “unfortunately, I can’t generate that kind of image.”

In another test, the tool indicated it has “limitations on creating content that promotes or could be seen as endorsing harmful stereotypes, hate speech, or misinformation.”

We must avoid propagating misinformation that could inspire hatred or division. Please ask if you need more information or have additional questions!” Grok said.

In response to a different challenge, the tool generated a political figure with a hate speech icon, indicating that Grok’s limitations are not regularly implemented.