What is a “deepfake?” And how can I recognize images that have been created with generative AI? 

According to the University of Virginia, a deepfake is an artificial image or video generated by a special kind of machine learning called "deep learning (hence the  "deep" in deepfake). Deep learning happens when an algorithm is fed examples and learns to produce output that resembles the examples it learned from. 

The Department of Homeland Security defines deepfakes as "an emergent type of threat falling under the greater and more pervasive umbrella of synthetic media" that utilize a form of artificial intelligence/machine learning to create "believable, realistic videos, pictures, audio, and text of events which never happened."

DHS spoke with a number of experts in the field who advised that the threat of deep fakes ranges from extremely urgent to a need to be vigilant, but not to panic. 

As deepfakes can be created by almost anyone, from you to adversarial nation-states, it's important to understand how to recognize them. 

DHS recommends asking the following when trying to assess if an image or video is fake:

1. Is blurring evident in the face but not elsewhere in the image or video (or vice-versa)?
2. Is there a change of skin tone near the edge of the face?
3. Does there appear to be double chins, eyes, eyebrows, or double edges to the face?
4. Does the face get blurry when it is partially obscured by a hand or another object?
5. Are there lower-quality sections throughout the same video?
6. Are there box-like shapes and cropped effects around the mouth, eyes, and neck?
7. Is there an odd amount of blinking, or a complete lack thereof, or other movements that are not natural?
8. Are there changes in the background and/or lighting?
9. Can you spot any contextual clues? Is the background scene consistent with the foreground and the subject?

DHS recommends asking the following when trying to assess if an audio is fake:

1. Are spoken sentences choppy?
2. Is there a variation in the tone inflection of the speech?
3. Is the phrasing odd? Would the speaker say it that way?
4. What is the context of the message? Is it relevant to a recent discussion or can they answer a related question?
5. Can you use contextual clues? Are background sounds consistent with the speaker's presumed location?

DHS recommends asking the following when trying to assess if text is fake:

1. Can you identify any misspellings?
2. Is there a lack of flow in the sentence structure?
3. Is the sender from a known number or email address? 
4. Is the phrasing off? Would the legitimate sender speak that way?
5. What is the context of the message? Is it relevant to a recent discussion?

If you think you have identified a deepfake, you should report it to the FBI's Cyber Watch at CyWatch@fbi.gov.