Deep fakes are expected to become a more prominent attack vector. Here’s how to identify them:
What is a deepfake?
Deepfake is the act of manipulating information by maliciously replacing real images and videos with fabricated ones. AI and ML are needed to create high-quality images, videos, and audio that can be used for deep fakes. Such use of AI, ML, and image replacement can be used to manipulate other types of information using less extreme manipulation techniques, such as misrepresenting information, isolating parts of information, or editing in a deceptive way. is different from Etay Maor, senior director of security strategy at Cato Networks, added: A human-like conversation bot
What do deepfakes look like?
Deep fakes come in many shapes and sizes. Some are simpler and some are more advanced. The most common examples of deep fakes are:
face swap
Face swapping is the act of replacing faces in a video or image from one person to another. Face swapping requires specialized software, but doesn’t have to be based on sophisticated technology. Today you can even find mobile apps that support face swapping. Face swapping available in mobile apps is typically limited to simple use cases, such as swapping a user’s photo with an actor’s face in a movie scene.
More advanced face swapping exists, but requires more model training and code, and consequently requires an expensive and resource-intensive GPU. An example of a more advanced face swap deep fake can be seen in this video. In this video, Tom Cruise is swapped with the presenter’s face.
This Tom Cruise face swap required two hours of GPU training and several days of professional video editing post-processing. While this may sound like a lot, it was considered an easier exchange than the others because the presenter has a haircut similar to Cruise’s and can impersonate his voice. This means less training and less post-processing.
Puppet Master (Lip Sync)
“Puppet Master”‘s deepfake is a technology that simulates the movement of a person’s mouth to make it look like they are saying something they are not actually saying. Compared to face swapping, which trains the model on the swapped new faces, the “puppet master” trains the model on the faces of the original images, especially the mouth movements.
Here’s what it looks like:
[warning – explicit language]
The technology behind “Puppet Master” is based on lip-syncing a mask, or composite of the original image, placed over a model of the person you’re impersonating.
audio
The third most prominent type of deep fake is voice-based. Audio deep fakes are audio files that take the voice of a real person and make it sound like they’re saying things they’ve never actually said. Audio deep fakes are created by taking audio files, assigning annotations to sounds, training an ML model based on the annotations to associate sounds with text, and generating new audio files.
Sounds are:
Deep fakes vs. cheap fixes
Not all altered images or sounds are deep fakes. Deep fakes are media synthesized or altered using AI, while cheap fixes are media synthesized or altered in low-tech ways that are easy to find. Often distorted and clearly manipulated. A cheap fix would be:
Deepfake cyber risk
Deep fakes are now more realistic, accessible, and faster to create than ever before. This makes it a powerful tool for weaponization. As a result, it poses risks to companies and countries. They can be used for cybercrime, social engineering, fraud, threat actor nation-state influence on foreign operations, and more.
For example, they used deep fakes to mimic the CEO’s voice to convince executives to transfer $243,000 into a fraudulent account. Etay Maor of Cato Networks, “Phishing attacks that make up business emails are becoming increasingly difficult to detect based on simple analysis of the language used. We need a holistic approach, one that can detect attacks at various chokepoints and not rely on isolated point products that are doomed to failure.” In another case, deep fakes were presented as evidence in a child custody case. I was.
Deepfakes can also be used to spread disinformation, the distribution of false information to influence public opinion or obscure the truth. For example, deep fakes can be used to impersonate world leaders to launch attacks or impersonate CEOs to manipulate a company’s stock price. In other cases, deepfakes enable plausible deniability, where people can deny all sources of media claiming to be deepfakes, causing a breach of social trust.
Finally, deepfakes can be used for defamation, or to damage someone’s reputation. For example, by creating revenge porn.
How to detect deep fakes
There are two main ways to accurately detect deep fakes.
- Low level detection method
- High level detection method
Low level detection method
Low-level detection methods rely on ML models trained to identify artifacts or pixelations introduced through the deep fake generation process. These artifacts may be invisible to the human eye, but models trained on real and deep fake images can detect them.
High level detection method
High-level detection methods use models that can identify semantically meaningful features. These include unnatural movements such as blinking, head poses, idiosyncratic mannerisms, and mismatched phonemes and visemes.
Today, these detection methods are considered accurate. However, as deep fake technology improves and becomes more sophisticated, it is expected to become less effective, requiring updates and improvements. In addition to these techniques, deep fakes can be detected by examining the media source of the received videos and images.
To learn more about different types of cybersecurity attacks and how to prevent them, check out Cato Networks’ Cyber Security Masterclass series.
