Even in its infancy, generative AI is fooling us on a daily basis. Some gags are harmless, like outlandish fashion on a red carpet, but others, like a fake call from President Joe Biden urging people not to vote, can shake our confidence in democratic institutions. With that in mind, McAfee has been working with Intel on a Deepfake Detector for audio fakes. Originally called Project Mockingbird, it made its debut at CES 2024, and we got a private demo this week at the security-focused RSA Conference in San Francisco.McAfee CTO Steve Grobman showed me a non-doctored video, with an AI-powered status bar showing the likelihood of fakery hovering near 0%. A second video, clearly fake, started off with a rating in the 60s, but as the clip went on it edged up to 95%. “Back in January when we announced the technology, it was mostly showcasing the AI itself,” Grobman told me at RSA. “Since then, we’re optimizing for the next generation of PC architecture.”The AI model runs on existing Meteor Lake CPUs, but adding the power of Intel’s NPU technology yielded a 300% performance gain. “We can run the inference engine in overlapping clips, which improves accuracy, and we can run multiple models at once,” said Grobman.Running on the PC instead of the cloud protects privacy. “We can analyze audio from any app on the system,” Grobman said. “But that could be very private. If it never leaves your device, we can guarantee privacy.” Plus, “by running the models locally, there’s almost no latency.”Focusing on audio at first “was very intentional,” according to Grobman, who noted that McAfee’s threat team found that many scam videos used real video with fake audio. Still, even if you have fake video, you almost invariably have fake audio. So “fake audio is a constant,” he said.
Recommended by Our Editors
I asked who the product is for. Will it be an app for the consumer, or a filter for the provider? Grobman said McAfee is currently developing technology for the consumer market, but isn’t making a product announcement. “We’re working on user experience and usability, to make sure that when we do interact with the user, they really understand what they’re seeing,” he said.I can certainly see how having a fake-detector running on my computer will be helpful.
Like What You’re Reading?
Sign up for SecurityWatch newsletter for our top privacy and security stories delivered right to your inbox.
This newsletter may contain advertising, deals, or affiliate links. Subscribing to a newsletter indicates your consent to our Terms of Use and Privacy Policy. You may unsubscribe from the newsletters at any time.
