Deepfake news this week.
What are deepfakes? Deepfakes are generally faked video and/or audio that are extraordinarily persuasive to the average viewer. They are generated using a combination of models including generative AI (GenAI) tools and competing models of Large Language Models (LLMs). AI has been good with images for a lot longer than with language, so the term "deepfake" emerged in 2017 when a Reddit moderator coined the term. It was used on a "subreddit" (a thread of discussion around a specific topic) for people who wanted to trade pornography deep fakes of celebrities, using the primitive face swapping software that existed at that time. For the technologically curious, it's interesting to note is that the way these images improve so quickly is they use two models. One for generating the images and another for detecting AI images, the two models provide continuous feedback to each other. So the tools for detecting deepfakes are co-opted into making deepfakes stronger.
(You might remember Insta and Snapchat filters that allowed you to swap faces with friends, the origin of that technology was much different than that harmless fun.)
Deepfakes are concerning for a lot of reasons, they are used in extortion, quick money grab scams, and in influencing politics and elections. Since the time of pornographic deepfake face-swapping, AI has improved on creating video, text and language, and, most famously in 2024, voice. In the 2024 election cycle, a deep fake audio voicemail went out encouraging democratic voters not to vote in this past election. This week it was the Republicans that got targeted.
An unknown hacker was able to deep fake Trump's Chief of Staff Susie Wiles and contacted members of Congress as her. In one case it appears that they may have created a print of her voice, like in the Biden case. Like "Signalgate" it's raised questions about the security in Trump's inner circles, because Wiles claimed her phone had been hacked, and that's how the individual was able to get access to her highest level contacts. The attempts sometimes seemed like garden variety "phishing" attempts to grab cash. At other times, it seemed politically motivated as one attempt requested that a Congressman make a list of all the people Trump might pardon.
In the case of the Biden administration, it was revealed that a Democrat worker had done it in order to show his team how easy it would be for a deepfake to cause chaos. It was "an inside job" as the saying gord. Steven Kramer paid a $150 commission to have someone create it. Prosecutors said the intent was not awareness but voter suppression, as he appears to have received a $250,000 payment for procuring the audio call and disseminating it from a New Hampshire contender for the presidency who would have benefitted from Biden supporters staying home.
Whatever the true intent, Kramer was fined $6m by the FCC and he faces 13 felony and 13 misdemeanor charges. In the case of Wiles, it's yet to be determined who is responsible and what their intent is. But it has called into question security around Trump's inner circle yet again.
Andy Greenberg
,
In other deepfake news Google's Veo 3 can make riots appear...
Andrew R. Chow
A deepfake of Pope Leo XIV went out that showed the first American pope praising the militant dictator of Burkina Faso...and a New Zealand MP, harkening back to 2017, held up a nude deepfake of herself during a meeting of Parliament, in order to demonstrate the shocking impact these tools can have on the social, cultural, and political sphere.
Singular XQ exists to provide open access research and low cost or pro bono training and career support for people with barriers to access in tech. Your subscriptions help us achieve our goals to upskill and educate the public and close the gap in technology. Consider upgrading today to help keep the public informed and educated free of special interest.
Help us keep educating the public free and low cost.
