KOHb - Getty Images
5 deepfake detection tools to protect enterprise users
Deepfakes are wreaking havoc worldwide -- and they're likely just getting started. To fight fire with fire, CISOs should consider AI-enabled deepfake detection tools.
As tools to create deepfakes become more sophisticated, affordable and accessible, CISOs should prepare for a growing number of social engineering attacks that hinge on fraudulent audio and video. In the coming months and years, many enterprise end users will likely encounter highly convincing deepfakes -- impersonations of colleagues or other trusted contacts, for example -- that could induce them to transfer funds or give sensitive information to threat actors.
In addition to educating users on evolving deepfake social engineering threats, security leaders should keep an eye on an emerging category of defensive AI and how it could help them protect their organizations: deepfake detection technology.
Let's look at five of the top tools that CISOs can use today to detect deepfake videos entering their organizations.
5 deepfake detection tools to consider
The deepfake detection market is expanding rapidly, with many more tools available than this article can cover. I selected the sampling of tools below, listed in alphabetical order, based on the following factors:
- Their ability to analyze videos without relying on video watermarking. Video editing software, such as Adobe and Canva, can add watermarking -- a logo or text overlaid on the video -- to denote ownership or copyright. This isn't a reliable indicator of deepfakes because watermarks can also be forged.
- Their use of a variety of deepfake detection techniques. Most tools today support only one or a few techniques, which means they tend to be better at identifying certain types of deepfake videos but might not be able to identify others at all.
Attestiv Video Platform
The Attestiv Video Platform works by uploading videos to the cloud and analyzing them for signs of editing, manipulation or AI generation. The software then rates the video's suspiciousness on a scale of 1 to 100.
Attestiv provides its software as both a web platform and an API.
Deepware Scanner
Deepware Scanner analyzes human faces in videos for indications -- such as distorted features or unnatural movements -- that they have been altered or switched, then reports the likelihood that the videos are deepfakes. It doesn't analyze any other video elements, such as audio tracks.
Deepware Scanner is currently free and available as a web platform and via API and SDK. Users can upload videos or submit URLs for scanning.
DuckDuckGoose's DeepDetector, Waver and Phocus
DuckDuckGoose provides the following tools for enterprise-grade, multimodal, real-time deepfake detection across live and recorded content:
- DeepDetector. Real-time detection of deepfake images and videos in on-premises deployments.
- Waver. Real-time, language-agnostic deepfake speech detection. Available via API integration.
- Phocus. Web platform that integrates DeepDetector and Waver for deepfake detection across images, videos and audio. Accessible via a browser.
DuckDuckGoose claims its tools have a deepfake detection accuracy of approximately 96%, with an analysis time of less than a second. The software reports the likelihood of each video being a deepfake and also explains the reasoning behind its estimate.
Reality Defender
Reality Defender scrutinizes files for AI-generation and tampering across audio, video and images, and monitors real-time video streams to detect impersonation attempts.
The company offers a variety of cloud and on-premises deployment options, including APIs and plugins for applications with real-time video streams, such as popular video-conferencing software.
Sensity AI
Sensity AI analyzes video, audio and images for signs of manipulated faces, synthetic voices and AI-generated images, while also reviewing the structures of the files themselves. Sensity claims its software has a deepfake detection accuracy of 98%. The platform also offers interactive training modules to help employees learn how to spot deepfakes.
Sensity AI is available for on-premises and cloud deployments, with a web platform and API access.
Karen Kent is the co-founder of Trusted Cyber Annex. She provides cybersecurity research and publication services to organizations and was formerly a senior computer scientist for NIST.
