Google’s AI video generator, Veo, is being used to create videos depicting racist and antisemitic tropes, which are then being shared to millions of followers on social media. Warning: some readers may find the video and images in this article offensive and disturbing.
An investigation by Media Matters for America, which monitors media outlets for conservative misinformation, found examples of videos in which black people were depicted as monkeys, criminals and obsessed with watermelons and fried chicken.
It found several examples of racist AI videos attracting millions of views on TikTok (see video below), but B&T has observed similar content being spread on other social media platforms, including Instagram and YouTube.
The discovery highlights the battle that social media companies face in monitoring and preventing highly powerful AI video generating tools from being weaponised to spread racism and hate on their platforms. Such content goes against their community standards and TikTok has already taken steps to remove the videos identified in the Media Matters report.
In one of the videos above, black people are depicted as monkeys and break into a KFC restaurant to steal chicken, while in another video a white woman screams that ‘I’m pregnant’ while her supposed partner, an ape, is seen driving away in a muscle car.
Another video shows a police officer on “low numbers” using a fishing rod and a watermelon bait to attract a black woman a video that was viewed 2.1 million times. Another video depicts Big Foot on horseback dressed in a Klu Klux Klan robe and hood chasing a black man.
There are videos of South Asian being chased by a giant soap and shower head, and Orthodox Jewish men chasing large gold coins down a hill, while another video depicts a Jewish man enjoying a concentration camp and saying that while it is a bit smokey, “something smells really nice out here”.
On TikTok, B&T found dozens of videos of police using watermelons to attract black people or monkeys called “Nikka” – presumably to bypass detection of a similar sounding racist term also starting with ‘N’. There were also dozens of videos portraying anti-semitic Jewish tropes.
All of the videos identified by Media Matters and observed by B&T are clearly watermarked ‘Veo’, and are shared across social media platforms.
Google’s Veo 3, which launched in late May to great fanfare, allows the general public to create eight second videos.
When using Veo a privacy policy disclaimer states, “labs.google/fx is an experimental technology and may sometimes give inaccurate information, including about people, or inappropriate information that doesn’t represent Google’s views”.
B&T contacted Google to find out more about Veo 3, its usage policies and what steps it has taken to prevent Veo from creating racist and hateful content. Google did not respond prior to this article being published.
TikTok has clear community guidelines that prohibit videos depicting hate speech and dehumanising content to racial and ethnic groups.
A TikTok spokesperson told B&T: “We proactively enforce robust rules against hateful speech and behaviour and have removed the accounts we identified in the report, many of which were already banned prior to the report publishing.”
