Musk’s AI chatbot Grok generates deepfakes of PM Modi and other illegal images

The Sways

3 months ago

Elon Musk’s AI chatbot Grok can be used to generate images of sexual assault, school shootouts, children’s TV characters going on killing sprees, and more, per general users and journalists

The AI chatbot Grok on Elon Musk-owned social media platform X is being used to generate graphic, explicit, fake, or illegal media soon after its image-generation capabilities were announced, claimed general users and journalists.

Other AI text-to-image generators generally use content filters or guardrails to block users from creating fake pictures of real political figures or celebrities, for safety reasons. This is especially key in the run-up to the U.S. presidential election towards the end of the year.

Just by entering a simple text prompt, The Hindu was able to use the new Grok 2 mini (beta) model to generate highly realistic deepfakes featuring Indian Prime Minister Narendra Modi as well as U.S. Vice President Kamala Harris wielding rifles.

Other X users shared graphic images they claimed were generated with Grok. These included a prominent American musician in a state of undress, Disney characters going on a killing spree, and even a gun-wielding Musk standing over dead or blood-drenched victims in a school.

Tech outlet The Verge also reported being able to create blatantly fake images of real life figures.

However, some basic controls and guardrails seemed to be in place, as Grok did not allow The Hindu to generate violent crime scene photos.

“I will not generate or describe that type of image. However, I can certainly help with other creative projects or discuss ways to convey intense scenes through writing or art that don’t involve graphic violence,” said Grok in response to a prompt to generate a blood-drenched active shooter scene for creative purposes.

Musk hailed the new model’s image generation capabilities and urged users to pay the monthly fee in order to access the chatbot.

Despite this, other X users claimed they were able to trick the model into showing children being harmed, or Disney and Nintendo characters carrying out adult activities.

AI startup xAI announced the Grok-2 beta release on August 14. Earlier versions of Grok have been criticised by regulators and digital safety advocates for spreading fake news and misinformation.