“Deep” in deepfake comes from deep learning, an AI branch that lets data go in, get analyzed, and make guesses like a human brain would. Until recently, most deepfakes came from Generative Adversarial Networks (GANs). GANs pit two neural nets against each other: one makes stuff, the other judges it. This back-and-forth leads to more lifelike results.
Generative AI’s boom sparked a rise in diffusion model deepfakes. While old school deepfakes just swapped celeb heads onto actor bodies or cooked up new dialogue, diffusion models build deepfakes from scratch. Cybersecurity experts are buzzing about how deepfakes are evolving. They reckon generative AI didn’t just make them look better, but it also slashed the cost of creating them.
Looking on the bright side, deepfakes are now also easier to detect, as companies race to develop deepfake detection tools, an emerging category of cybersecurity technology.
Why deepfakes?
People now use deepfakes to rig elections, steal identities, commit fraud, and spread lies. For example, in May 2023, a convincing fake photo of the Pentagon on fire caused stock prices to drop. US officials jumped in to clarify that no blast actually took place and the photo was a fake. In another situation covered by CNN News, a fake video of a chief financial officer tricked a finance worker, leading to a mistake that cost $25 million.
Recent incidents show how deepfakes are impacting the 2024 US Presidential Election. CNBC News reported that in New Hampshire, a fake video showed President Joe Biden telling voters to skip the presidential primary. Also, NBC News reported on fake videos on X that made it seem like Taylor Swift supported Donald Trump. This underlines the increase in fake videos as the elections draw near.
Deepfakes threaten online content authenticity. The human eye struggles to spot the difference between real and fake material. Bad actors can spread lies, hurt reputations, and cause conflict with fake images, videos, and audio that have the power to shape public opinion on a massive scale. Fortunately, AI advances have a bright side too. New tools can now detect deepfakes with impressive accuracy.
What is a Deepfake Detection Tool?
A deepfake detection tool is a piece of software designed to spot fake videos or images. It looks at digital content in different ways to figure out if someone has messed with it or made it using AI.
Experts predict the global market for deepfake detection software will boom. They expect it to grow by 38.8% each year from 2024 to 2029. The video deepfake detection segment is expected to hold the largest share of the market over the forecast period. The deepfake image detection market will also explode, jumping from $0.6 billion in 2024 to $3.9 billion by 2029. That’s a whopping 41.6% increase every year during that time.
Tools to Detect Deepfakes
Deepfakes are getting better, but so are the tools to spot them. Many of them are free, enabling security experts and regular users alike to protect against fake digital content. Let’s take a look at some great free tools to find deepfakes.
Deepware
Deepware is a sophisticated program that utilizes AI and machine teaching methods to find and stop deepfakes. It recognizes videos, pictures, and audio tracks and figures out if they are real or fake. It’s very easy to use and offers a special scanner feature that lets users look at certain parts of visual and sound messages by just typing in URLs.
An artificial intelligence model examines the uploaded content to check if it’s real and to spot any tampering or fake parts. This service detects deepfakes in real time for all users, goes through videos on different platforms, and checks if they are authentic before they get shared or published.
True Media
True Media is a non-profit organization committed to fighting AI-based disinformation. It has launched a fake detection technology for reporters to use ahead of the 2024 US election, but it can be applied to deepfake content from anywhere.
The free tool is a valuable resource to government officials, analysts, campaign staffers, universities, nonprofits, and reporters from news organizations.
WeVerify
WeVerify tackles complex challenges in verifying content by using a team-based method, free AI tools, and easy-to-understand visuals. It checks social media and internet posts for wrong information, analyses it in a larger context, and reveals false and made-up stories. This is done by examining facts and using a public, blockchain-based record of known untrue stories.
WeVerify also features the Google Chrome add-on “InVID WeVerify.” This tool is designed to aid journalists, advocates for human rights, and other users in saving time and boosting their ability to check facts on social platforms with multimedia like videos and photos.
This toolkit can do reverse image searches on several search engines, look into the database of known manipulated material, read video and image details, check video copyrights, and analyze content from different platforms like Facebook, Instagram, and YouTube (though it can work only partially on X because of API policy changes).
It also has features to spot altered photos. Moreover, as a part of the EnVisu4 IFCN project, the CheckGIF module lets users make GIFs that show the differences between altered images and the original ones.
Other options
Other tools specializing in deepfake identification for a fee include Sentinel, Sensity, Deepfake Detector (audio-video material under 5MB can be checked for free), Bio ID (a demo version can be accessed after registering on the platform), and Oz Forensics.
Giants such as Google, Intel and Microsoft have also created tools to detect deepfakes. However, Google’s SynthID, Intel FakeCatcher and Microsoft Video Authenticator are not standalone commercial solutions. These technologies are integrated into various products and services offered to individuals, companies or research institutions.
Conclusion
Deepfake detection is an emerging category of cybersecurity technology. Many of these tools are still in beta or early stages of development, which means that their feature sets are still evolving and may vary between different solutions. However, there are some interesting features that, combined with human training, should help in the battle against deepfakes.
