Earlier this year, a Hong Kong bank employee wired nearly $26 million to local bank accounts after being told to do so by the bank’s chief financial officer on a Zoom call. The trouble is that the CFO and other people the employee recognized on the call were digital doppelgangers – so-called deepfakes of the real people. By the time the employee figured it out, the money was gone.
The development of deepfakes has been a gradual process spanning nearly two decades. But advances in computational power, data availability, and algorithm design have made them real threats rather than the novelties they once were. Instead of funny videos of a fake Tom Cruise or more malicious efforts swapping out porn actress faces for those of famous females, the technology is getting good enough to cause broader social harm.
As with so much of cybersecurity, detecting deepfakes is a cat and mouse game between generative and discriminative algorithms. While there is hope that discriminative models for detecting fakes may progress faster than generative ones, keeping up with the rapid advancements in generative AI will be an ongoing struggle. Generative algorithms are always much harder to build than discriminative algorithms.
While the algorithms have achieved detection accuracies as high as 90 percent on benchmark datasets, they are not foolproof and may struggle with carefully crafted deepfakes that have undergone post-processing. Lyu acknowledges, however, that audio deepfakes are currently the most challenging type of synthetic media to detect.
Source: News Formal (newsformal.com)
United States Latest News, United States Headlines
Similar News:You can also read news stories similar to this one that we have collected from other news sources.
Source: ForbesTech - 🏆 318. / 59 Read more »
Source: washingtonpost - 🏆 95. / 72 Read more »
Source: ScienceDaily - 🏆 452. / 53 Read more »
Source: CalMatters - 🏆 261. / 63 Read more »
Source: verge - 🏆 94. / 67 Read more »
Source: IntEngineering - 🏆 287. / 63 Read more »