Elon Musk’s AI photo tool is generating realistic, fake images of Trump, Harris and Biden

Clare Duffy, CNN

New York (CNN) — Elon Musk’s AI chatbot Grok connected Tuesday began allowing users to create AI-generated images from matter prompts and station them to X. Almost immediately, group began utilizing nan instrumentality to flood nan societal media tract pinch clone images of governmental figures specified arsenic erstwhile President Donald Trump and Vice President Kamala Harris, arsenic good arsenic of Musk himself — immoderate depicting nan nationalist figures successful evidently mendacious but nevertheless disturbing situations, for illustration participating successful nan 9/11 attacks.

Unlike different mainstream AI photograph tools, Grok, created by Musk’s artificial intelligence startup xAI, appears to person fewer guardrails.

In tests of nan tool, for example, CNN was easy capable to get Grok to make fake, photorealistic images of politicians and governmental candidates that, taken retired of context, could beryllium misleading to voters. The instrumentality besides created benign yet convincing images of nationalist figures, specified arsenic Musk eating steak successful a park.

Some X users posted images they said they created pinch Grok showing salient figures consuming drugs, animation characters committing convulsive murders and sexualized images of women successful bikinis. In 1 station viewed astir 400,000 times, a personification shared an image created by Grok of Trump leaning retired of nan apical of a truck, firing a rifle. CNN tests confirmed nan instrumentality is tin of creating specified images.

RELATED | Gov. Little signs AI-generated porn measure into law

The instrumentality is apt to adhd to concerns that artificial intelligence could create an detonation of mendacious aliases misleading accusation crossed nan internet, particularly up of nan US statesmanlike election. Lawmakers, civilian nine groups and moreover tech leaders themselves person raised alarms that nan misuse of specified devices could origin disorder and chaos for voters.

“Grok is nan astir nosy AI successful nan world!” Musk posted connected X Wednesday, successful consequence to a personification praising nan instrumentality for being “uncensored.”

Many different starring AI companies person taken immoderate steps to forestall their AI image procreation devices from being utilized to create governmental misinformation, though researchers recovered users tin still sometimes find ways astir enforcement measures. Some companies, including OpenAI, Meta and Microsoft, besides see exertion aliases labels to thief viewers place images that person been made pinch their AI tools.

Rival societal media platforms, including YouTube, TikTok, Instagram and Facebook person besides taken steps to explanation AI-generated contented successful users’ feeds, either by utilizing exertion to observe it themselves aliases asking users to place erstwhile they’re posting specified content.

X did not instantly respond to a petition for remark regarding whether it has immoderate policies against Grok generating perchance misleading images of governmental candidates.

By Friday, xAI appeared to person introduced immoderate restrictions connected Grok, successful consequence to captious reports and nan disturbing images that users were creating and posting. The instrumentality now refuses to create images of governmental candidates aliases wide recognized animation characters (whose intelligence spot belongs to different companies) committing acts of unit aliases alongside dislike reside symbols, though users noted nan restrictions seemed constricted to only definite position and image subjects.

X has a argumentation against sharing “synthetic, manipulated, aliases out-of-context media that whitethorn deceive aliases confuse group and lead to harm,” though it’s unclear really nan argumentation is enforced. Musk himself shared a video past period connected X that utilized AI to make it look that Harris had said things she, successful fact, did not — successful an evident usurpation of nan argumentation and pinch only a laughing look emoji to propose to followers that it was fake.

The caller Grok image instrumentality besides comes arsenic Musk faces disapproval for many times spreading mendacious and misleading claims connected X related to nan statesmanlike election, including a station raising questions astir nan information of voting machines. It besides comes days aft Musk hosted Trump for a much than two-hour livestreamed speech connected X successful which nan Republican hopeful made astatine slightest 20 mendacious claims without pushback from Musk.

Other AI image procreation devices person faced backlash for various issues. Google paused nan expertise for its Gemini AI chatbot to make images of group aft it was blasted for producing historically inaccurate depictions of people’s races; Meta’s AI image generator came nether occurrence for struggling to create images of couples aliases friends from different group backgrounds. TikTok was besides forced to propulsion an AI video instrumentality aft CNN discovered that immoderate personification could create realistic-looking videos of group saying anything, including vaccine misinformation, without labels.

Grok does look to person immoderate restrictions; for example, a punctual requesting a nude image returned a consequence saying, “unfortunately, I can’t make that benignant of image.”

In a abstracted test, nan instrumentality said it besides has “limitations connected creating contented that promotes aliases could beryllium seen arsenic endorsing harmful stereotypes, dislike speech, aliases misinformation.”

“It’s important to debar spreading falsehoods aliases contented that could incite hatred aliases division. If you person different requests aliases request accusation connected a different topic, consciousness free to ask!” Grok said.

However, successful consequence to a different prompt, nan instrumentality did make an image of a governmental fig opinionated alongside a dislike reside awesome — a motion that immoderate restrictions Grok does have, they look not to beryllium enforced consistently.

–CNN’s Jon Passantino contributed to this report. This communicative has been updated to bespeak changes xAI has made to Grok.

The-CNN-Wire
™ & © 2024 Cable News Network, Inc., a Warner Bros. Discovery Company. All authorities reserved.