Zurich, Switzerland, June 19th, 2024– LatticeFlow AI, the leading platform for building trustworthy AI applications, announces LatticeFlow AI Audio, the first product designed to find and identify model errors in audio AI applications in addition to the company’s already existing product for computer vision.
The timely release addresses the pressing demand for reliable voice deepfake detection solutions in advance of the 2024 U.S. elections. In this context, while speech is the most common and significant type of deepfake affecting electoral processes, the release of LatticeFlow AI Audio, combined with the company’s existing capabilities in computer vision, makes their platform the most advanced solution for deepfake detectors to combat sophisticated deepfake threats combining audio and video.
AI Against AI: An Escalating Race Between Deepfake Generators and Detectors
In the past 18 months, generative AI has democratized the threat landscape, enabling bad actors to easily craft realistic content without requiring complex skills or expensive equipment. AI deepfake detection solutions are now increasingly important, as they determine whether an audio and/or video segment is real or fake, essentially using AI to fight AI. However, as a result, it has effectively sparked a race between ever-increasingly sophisticated deepfake AI generation models and the detection models designed to counter them.
“We believe that there is no place for deepfakes in civil, ethical societies,” said Dr. Petar Tsankov, Co-founder and CEO at LatticeFlow AI. “Ethical, safe, and trustworthy AI are the foundational values of our company – and why we want to help address deepfakes detection in advance of the 2024 elections. LatticeFlow AI Audio is the industry’s only product that can automatically find and identify AI audio data and model errors, enabling our clients to rapidly develop more accurate AI deepfake detection applications”.
“The announcement of LatticeFlow AI’s technology in combating audio deepfakes, in addition to their already existing technology improving video deepfake detectors, is appreciated by journalists such as myself”, stated Chris O’Brien, renowned independent journalist. Prior to relocating to Paris, O’Brien covered Silicon Valley for The San Jose Mercury News and the Los Angeles Times. O’Brien added: “I am thrilled to see companies like LatticeFlow AI developing cutting-edge technology to combat deepfakes. Their innovation is a vital step towards preserving the integrity and accuracy of the information we often rely on as journalists.”
LatticeFlow AI Audio Finds and Fixes Critical Deepfake Detector Vulnerabilities In Under a Minute
LatticeFlow AI’s research team analyzed a best-in-class, publicly available voice deepfake detector. While the company boasts that its model has an impressive 99 percent accuracy of detecting deepfakes, this was not accurate across a broad range of voice samples that LatticeFlow AI tested, which had an 20x increase of detectors’ error rate. That is: 20% of deepfake speech samples were able to bypass the detector.
LatticeFlow AI Audio was able to fix this vulnerability in under a minute, creating a custom guardrail to effectively identify and block future attacks attempting to exploit this vulnerability.
Product Use Beyond Improving Deepfake Detectors
LatticeFlow AI Audio enhances the reliability of deepfake detection applications and supports a wide range of audio and speech AI applications, including both audio and video deepfakes.
This aligns with the global trend towards developing more complex multi-modal AI applications that integrate both vision and audio to deliver trustworthy solutions in an array of use cases. For example, in manufacturing, quality control applications increasingly rely on ultrasound sensors to improve accuracy in detecting hard-to-spot or even hidden defects. Similarly, remote identification verification solutions incorporate multiple sensors and defenses to counteract the growing capabilities of attackers attempting to spoof user identities.
About LatticeFlow AI
LatticeFlow AI is an award-winning ETH Zurich spin-off founded in 2020 by a globally recognized team of AI experts whose mission is to help customers automate their AI lifecycle. To achieve this goal, the company has built the first scalable platform that guides AI teams through the AI lifecycle by automatically finding and fixing data and model issues. LatticeFlow AI’s customers range from the US Army and the Singaporean Defense, Fortune 500 companies such as Siemens Mobility, to AI scaleups such as Actuate, Athena AI, and Fyusion. The company was featured on the prestigious CB Insights AI100 List of Most Innovative AI companies in 2022 and 2023, and also won the Swiss AI Award in 2022 and the US Army Global AI Award in 2021.
