Grok-2, the latest version of the chatbot from Musk’s xAI startup, is available to X Premium subscribers. Let users make images that violate content restrictions used by similar tools. As a result, bizarre and occasionally disturbing images depicting real people—including Musk himself — were shared on X.
Users have shared images mocking his relationship with Donald Trump. Some show Musk on all fours as the former president walks him on a leash. Others depict Musk kissing Trump (who is 78, you know), while one shows the Tesla boss holding a sign: “I AM A PEDO!”
More alarmingly, one user claimed to have found a way to make Grok generate images of school shootings, posting examples depicting Musk holding a rifle in a classroom that has bodies and pools of blood on the floor.
Another showed Musk raising a gun above his head outside a building with the words “Uvalde Elementary School” above its doors—a reference to the 2022 school shooting in Texas in which 19 children were killed. Others show an armed Musk standing outside a school alongside Mickey Mouse, wearing a top bearing the words “crisis actor.”
But Musk wasn’t the only person targeted in the chaos. While some AI image-generators won’t allow their tools to make fake images of politicians with the election approaching, Grok doesn’t appear to have similar guardrails.
Users were, therefore, able to share their depictions of Trump and Vice President Kamala Harris having a baby together or carrying out the 11 September 2001 terror attacks. They could also cook up images of former presidents Barack Obama and George W. Bush taking cocaine.
Concerns about AI’s use to create and spread political misinformation have been growing. Rapidly advancing technology makes it increasingly difficult, in some cases, to distinguish fiction from reality. Trump himself contributed to a viral conspiracy theory over the weekend by falsely suggesting that images of a crowd at a Harris rally were made with AI.
Calls to moderate image generation software have also centred on the technology’s use in making deepfake porn. In January, X made Taylor Swift unsearchable on its platform after explicit AI-generated images of the singer circulated, a phenomenon which both SAG-AFTRA and the White House denounced.
Those images appeared to have come from a site notorious for making fake nude images of celebrities. According to the Guardian, Grok will refuse to create fully nude photos. Nevertheless, The newspaper claimed to have reviewed prompts and images depicting Swift, Harris, and Rep. Alexandria Ocasio-Cortez (D-NY) in lingerie.
The report also found that Grok would create images showing Minnie Mouse—a copyrighted character—when prompted in a certain way. The tool was also used to produce images of “Mickey Mouse saluting Adolf Hitler and Donald Duck using heroin,” according to the Guardian. X users also used Grok to make images featuring Nintendo characters.
“Grok is the most fun AI in the world!” Musk wrote Wednesday in response to a user praising the new version of the tool as the most “uncensored model of its class yet.”
We are beginning to think that a civil war is inevitable at X and Tesla.
