After Russia invaded Ukraine, hackers put a one-minute video on Ukrainian news websites claiming to show Ukrainian president Volodymyr Zelenskyy telling soldiers to surrender. We do not know exactly who made the video, but we do know one thing: It was a deepfake. It was not well-made and contained several obvious signs of being fake. However, the video still went viral before social media companies removed it from their platforms.
This was not the first deepfake that fooled people and while it was easy to spot, technology is rapidly evolving, so it’s important to know what deepfakes are and why they are so dangerous so that you can spot them when one inevitably comes across your social media feed. Here’s everything you need to know about deepfakes.
How do deepfakes work?
Deepfakes are videos, audio, or images meant to look like a person said or did something that they didn’t. They are created by feeding the existing media of a person into software programs that use a kind of artificial intelligence (AI) called “deep learning.” The AI learns key traits of its subjects, such as facial features, mannerisms, or the way the person speaks, and it learns the traits of other people, too. It uses all this data to create new or manipulated images, videos, or audio. AI can also create pictures of people who do not actually exist through what is called generative adversarial networks (GAN).
Deepfakes can be made quickly and with free and easy-to-use software, but the quality of these deepfakes tends to be low and, therefore, easy to spot. For example, there are playful phone apps that use face-swapping technology to quickly put a person’s face on another person’s body in a video. More convincing deepfakes can take longer to make and require more sophisticated, expensive editing software and computer equipment.
What’s the difference between deepfakes and cheapfakes?
Cheapfakes are videos, images, or audio that have been edited, usually by using low-quality software or basic editing tools. As of now, cheapfakes are the main type of fake or manipulated media content you will see online. Cheapfakes include things like edited images, videos of separate events stitched together to make it look like one, or editing out or slowing down words by a speaker to change the meaning of what they said. For example, an edited video of President Joe Biden’s July 4 speech was slowed down to make it sound like he was slurring his words. In fact, ahead of the 2020 election, several videos posted on social media of Biden were cut short and posted out of context to promote the false claim that he had cognitive disabilities because of his age.
Are deepfakes illegal?
Currently, there is no federal law outlawing deepfakes in the US and only a handful of states have passed legislation that regulates them. Matthew F. Ferraro, a former intelligence officer and an attorney who specializes in deepfakes, says that most of those state laws primarily ban nonconsensual deepfake pornography and deepfakes of political candidates before an election. New York and Louisiana give people property rights over how their “own digital likeness” is used, and Ferraro says it would be illegal there to create a deepfake of a celebrity telling people to buy something you sell.
More states are considering new laws regulating deepfakes. “Not only is creating and sharing nonconsensual deepfake pornography morally wrong, but it’s also illegal in many states,” Ferraro warns. “Everyone should know that making and distributing those kinds of images isn’t funny and can get them in really hot water.”
How can you spot deepfake videos and photos?
As the technology gets better, deepfakes will become harder to detect. However, there are a number of ways you can spot them. First, check for odd-looking body proportions that seem off or differences in a person’s voice or accent. In the Zelenskyy deepfake, the president’s head looked like it had been pasted onto his body and his body did not move throughout the video. AI-generated pictures of people who are not real often have abnormalities you can spot on their ears, hands, eyes, and hair.
You can also look for the original image, video, or audio clip first posted online, either by searching on news sites or conducting a reverse-image search where you upload a screenshot into a search engine like Google. When you find the original, compare it to what you saw or heard.
Lastly, deepfakes are often generated to make a big claim, especially when the subject is a politician or public figure. Go to news sites and fact-checking organizations to verify if what you are seeing has been reported in the news or has been debunked as a deepfake or cheapfake by organizations like Poynter.
What’s the most common use for deepfakes?
Deepfakes have been created to impersonate celebrities like Tom Cruise, to scam people out of money, to impersonate public figures like Facebook CEO Mark Zuckerberg, and to run fake social media accounts using AI-generated profile pictures. However, the most common deepfakes are of nonconsensual pornography — that is, fake pornographic videos or images made without the consent of the person in them. Since December 2018, Sensity AI, a tech company that tracks deepfakes, found that 90 to 95% of deepfakes are nonconsensual porn and, as of 2021, 90% of those are of women, as reported by MIT Technology Review.
The first known deepfake videos, according to CNN, appeared on Reddit in 2017 and put the faces of celebrities on the bodies of porn stars. Some websites have tried to profit by selling access to deepfake porn videos. For example, one website took photos of women posted online and used AI to remove the clothing, turning them into nude images.
Why are deepfakes potentially dangerous?
Deepfakes increasingly pose a threat to national security because of their ability to trick people. A deepfake video, audio, or image could be used to make false claims about the integrity of an election or pretend to show a political leader making an incendiary comment, like threatening to attack another country. In the deepfake of Zelenskyy, the entire tide of the war could have changed if Ukrainian soldiers had believed it really was him and put down their weapons, or if it had convinced Ukraine’s allies to stop sending weapons and supplies.
Ferraro says a hyperrealistic deepfake video of Biden declaring war on a country like North Korea could create a “nightmare scenario,” especially “during a period of heightened tensions on the Korean peninsula.” In the worst possible outcome, he says, such a video could “cause confusion and prompt the North Korean regime…into launching an attack on Seoul, South Korea.” This is the kind of scenario where a deepfake video has the potential to lead to mass casualties and threaten international security.
Stay up-to-date with the politics team. Sign up for the Teen Vogue Take
