Unmasking the Fakes: Emerging Tools to Identify Deepfakes

In recent years, the rapid advancement of AI and machine learning has given rise to a new form of synthetic media known as deepfakes. These highly realistic manipulated videos, images, audio or text files have the potential to deceive and mislead, posing significant challenges to trust and authenticity in the digital age. As deepfakes become more common and sophisticated, the need for reliable detection tools has never been greater. In this article, we will explore some of the leading deepfake detection tools currently available.

What harm can deepfakes cause us?

1. Security Threats

Deepfakes can pose security risks by impersonating certain individuals, leading to fraud, espionage, or unwarranted access to confidential information. For example, in Feb 2024, CNN reported that a finance worker at a multinational firm was tricked into paying out $25 million to fraudsters using deepfake technology to pose as the company’s chief financial officer in a video conference call.

2. Misinformation and Fake News

Deepfakes have the potential to spread misinformation rapidly by fabricating realistic videos or audio recordings. This can manipulate public opinion, influence elections, and undermine trust in media. Cybercrimes using Deepfakes have widespread impacts beyond individual victims, affecting political landscapes, stock markets, and public trust. Check out the video below for how AI might impact elections.

3. Privacy and Consent Violations

Deepfakes can be used to create convincing and unauthorized content of individuals, infringing on their privacy and consent. This misuse can lead to significant personal and professional harm. For example, in January 2024, fake nude images of Taylor Swift spread on X (formerly Twitter), Meta, and Telegram. These images, showing explicit and sometimes violent images without her permission, reached hundreds of millions of people.

Our hope - Deepfake detection tools!

In light of the increasing prevalence of malicious deepfakes, both researchers and technology companies have been working on creating a range of tools designed to identify these manipulated media. Here are some of the most prominent deepfake detection tools that are currently accessible. By familiarizing yourself with this information, you will have a better understanding about the progress and implementation of deepfake detection technologies.

1. Reality Defender

Reality Defender is an enterprise-level solution for detecting deepfakes in video, audio, image and text content. Reality Defender's tool analyzes the content provided and generates a probability score of possible manipulation within the content.

Reality Defender listed a number of case studies where they helped prominent organizations detect dangerous deepfakes. Below are a few examples.

Please note that the links below only shows a summary of the case studies. If you'd like to read the full case studies, Reality Defender website will ask you to download the report.

• Reality Defender helped a senior government official from Canada detect and disprove deepfake videos that falsely depicted them in a negative light.
• Reality Defender helped an Asian national public broadcaster maintain journalistic integrity and audience trust by identifying and preventing the spread of manipulated content.
• Reality Defender helped a Tier-one bank detect deepfaked audio to deceive employees.

Read more case studies here on Reality Defender's website if you are interested. Pretty interesting to read. I highly recommend.

2. Microsoft Video Authenticator

Microsoft has developed a powerful deepfake detection tool called Video Authenticator. This tool analyzes videos and images to determine whether they have been artificially manipulated. It uses a combination of AI algorithms and digital forensic techniques to detect inconsistencies and artifacts that are often present in deepfakes. Video Authenticator also provides a confidence score indicating the likelihood of manipulation, helping users make informed decisions about the authenticity of the media they encounter.

Check out the tweet below that illustrates how Microsoft's Video Authenticator spots deepfakes. The Tweet is a bit outdated (I was not able to find a newer one unfortunately), but this tweet should be able to give you an idea how it works.

3. Sensity AI (formerly Deeptrace Labs)

Sensity, previously known as Deeptrace Labs, is another leading provider of deepfake detection solutions. Their platform combines cutting-edge AI algorithms with expert human analysis to identify and combat the spread of Deepfakes. Sensity AI offers a range of services, including deepfake detection, threat intelligence, and content authentication. They work closely with organizations across various industries to help them safeguard against the risks posed by deepfakes. Check out the tweets below as an example.

4. Google SynthID

Developed by Google’s AI research lab DeepMind, SynthID is a watermarking tool designed for AI-generated content. Currently it's beta-launched. SynthID is a tool for watermarking and identifying AI-generated images or audio files. The way it works is that when users create images using Imagen – one of Google's latest text-to-image models, they can embed a digital watermark undetectable to human eye. When these images are uploaded into SynthID, the watermark can be detected by SynthID. Even though this tool is currently only applicable for Google's AI-generated audio or image files, but in the future it could be expanded for use across other AI models.

5. Intel FakeCatcher

Intel FakeCatcher is a cloud-based AI deepfake detection tool developed by Intel. FakeCatcher claims to be able to accurately detect fake videos 96% of the time. The technology analyzes tiny changes in pixel color caused by blood flow in a video. It then compares the patterns across multiple frames and uses an algorithm to determine if the video is genuine or fabricated. The Tweet below explains the technology behind FakeCatcher well.

Given that the field of deepfake detection is still a new area, there are many other players this article was not able to cover. I plan to write more about this field in the near future.

Looking Ahead...

As the threat of deepfakes continues to grow, the development and adoption of reliable detection tools become increasingly critical. The tools mentioned in this article represent some of the leading efforts in the fight against deepfakes. However, it is important to recognize that the battle against deepfakes is an ongoing one. As detection methods improve, so too will the sophistication of deepfake technology. Staying vigilant, educated, and proactive in the face of this evolving threat is essential to maintaining trust and authenticity in the digital realm.