What are Deepfakes and How do I Spot One?

What is a Deepfake?

A deepfake is a fraudulent piece of content—typically audio or video—that has been manipulated or created using artificial intelligence. This content replaces a real person’s voice, image, or both with similar looking and sounding artificial likenesses.

Think of a deepfake video like an image that’s been heavily edited. If you took photos at the beach but wanted to remove aspects of the image, like trash or other beachgoers, and add in other aspects, maybe an extra palm tree or a parrot, someone who is skilled at photo manipulation could do that for you. It would still be your picture of the beach, but it wouldn’t be true to life. A deepfake video is composed of little bits of truth but put together in a way that makes it false.

In some deepfakes, videos of real people are doctored to make it seem like they’re saying things they’d never say in real life. In others, one person’s face is superimposed on another person’s body, convincingly making it seem like they’re physically doing something they wouldn’t normally do. One example of this is a video shown during a lecture about deep learning at MIT that showed a former president welcoming students and explaining some aspects of a college course. The entire speech and video were created with deepfake technology. While that video was created to showcase the technology that students would be learning about, it frequently has more nefarious uses. It’s not uncommon for female celebrities to learn that their faces have been superimposed onto other women’s bodies in pornographic films, due to deepfake technology.

Deepfakes are frequently used to spread disinformation, and can be used in scams, election manipulation, social engineering attacks and other kinds of fraud.

Microsoft Defender

Stay safer online with one easy-to-use app<sup>1

1</sup>Microsoft 365 Personal or Family subscription required; app available as separate download

Learn more

How does Deepfake Technology Work?

Artificial intelligence and machine learning play a huge role in creating a deepfake. Computer systems called artificial neural networks are loosely based on the human brain and are built to recognize patterns that exist in data.

To make a deepfake video, a creator needs to first supply the artificial neural network with real video footage to teach the AI to identify someone’s physical characteristics. The AI learns the person’s expressions through watching hundreds of hours of video and learns what a person looks like from all angles.

Next, the neural network that has been “trained” to identify a person’s face is combined with graphics techniques that overlay real footage of a person with the speech and facial patterns that have been learned by the “trained” network. The network realizes the similarities between the two faces and swaps the characteristics like ears and eyebrows accordingly.

Another way that deepfakes are created is through a system called a Generative Adversarial Network, or GAN. This network analyzes millions of existing faces to recognize patterns, like where features like eyes and ears are located, and can use this information to produce faces that don’t actually exist but look real. This same network can also use its database to create fake photos of real people, and then create a video out of those fake photos.

Are Deepfakes Dangerous?

Because these videos and photos are created with deception in mind, they can be very dangerous when used maliciously. Around the world, deepfakes have been embroiled in political controversy, they have been used to stir up political unrest and even led to an attempted military coup in the African country of Gabon.

Deepfakes make misinformation seem legitimate, because the created videos can make it seem like it’s coming from a credible source. This can wreak havoc on elections; a deepfake video of a candidate saying racist things would sabotage their chances. The average voter might not understand that it was a fake and even if the candidate could prove that it wasn’t true, it would have had a negative impact on their public perception.

Similarly, people behaving badly might simply deflect when caught and say that any incriminating photos or videos are fakes. This kind of synthetic media has a goal of a society where people cannot distinguish truth from lies, especially in regard to the news. When that trust has been eroded, it’s easier to raise doubts about the legitimacy of events.

As this technology because more sophisticated and accessible, deepfakes could cause trouble in the court system, where videos of faked events might be presented as evidence. The capacity for copying biometric data means that scammers have the potential to use deepfakes to trick unsuspecting people out of their money or personal information. A scammer could use a deepfake to videocall an elderly person and pose as a grandchild asking for money, and easily dupe someone.

How to Spot Deepfakes and Protect Yourself

A poorly made deepfake photo or video may be easy to identify, but detection might prove a little more difficult on a professionally made video. There are a few characteristics that might help identify them as being fake:

• Strange eye movements and a lack of blinking
• Unnatural hair or hair movement
• Patchy skin coloration.
• Strange positioning of hands, hands, and body
• Poor lip-syncing
• Strange lighting or obvious discoloration
• Voices that sound robotic and unnatural speech patterns

On a personal level, if you see a video online that seems totally unbelievable, it probably is. Do a little research and confirm that the source is reputable before deciding. If someone you know appears to call you and asks for something completely out of the ordinary, like money or an account number, double check with that person before acting.