Deepfakes are booming online, but South Africans struggle to spot them

2024-01-22 -

In a recent survey by cybersecurity firm Kaspersky, only one in five South Africans correctly identified an image generated using artificial intelligence. • This comes amid a 1 200% growth in the frequency of deepfakes in South Africa between 2022 and 2023, according to an identification company. • Prominent South Africans, including Elon Musk, Cyril Ramaphosa and Glynnis Breytenbach, were targeted by deepfakes last year. Images and videos made using sophisticated artificial intelligence tools are proliferating online and South Africans are struggling to spot them. A mere 21% of South African employees were able to tell the difference between a real image of a US actor and one generated by an artificial intelligence tool in a survey conducted by Kaspersky, a global cybersecurity firm. Double the number of respondents in the Kaspersky Business Digitization survey - or 42% of South Africans - were confident that they would be able to tell the difference between the images. Dmitry Anikin, a senior data scientist at Kaspersky, said that it is common for employees to overstate their digital skills. `Even though many employees claimed that they could spot a deepfake, our research showed that only half of them could actually do it,` said Anikin. A deepfake is a hyper-realistic video, image, or audio recording created using sophisticated artificial intelligence tools that often feature individuals saying or doing something that never actually happened. AI-generated misinformation deemed greatest threat in 2024 election year Deepfakes can be used in ploys to scam people out of money or as a means to influence public sentiment by spreading misinformation. There have already been high-profile South African incidents of deepfakes being used for both of these purposes. A deepfake video was released late last year that purported to show an SABC broadcast where the presenter introduces a new investment scheme launched by Elon Musk. The fake broadcast cuts to a digitally altered clip of Musk where he talks about the new algorithm he has developed that yields a R30 000 monthly return on an initial R4 700 investment. The video amassed over 130 000 views on YouTube in the three months since it was published

This was not the only video made in a similar style. The SABC released a statement and produced its own video explaining that those were fake. Politically, the DA`s Glynnis Breytenbach and President Cyril Ramaphosa were both targeted by deepfakes last year. In a deepfake video of Ramaphosa released in April, he is also shown in an apparent news broadcast presenting a plan to tear down the Voortrekker Monument and the Loftus Versfeld rugby stadium to create space for large diesel generators. In the deepfake targeting Breytenbach, her voice was recreated in a sound clip about loans that DA leader John Steenhuisen had seemingly accepted from the Bezos Foundation and a US desire to control the Cape`s shipping routes. 2023 was an inflection point for the emergence of these scams in South Africa. There was a 1 200% increase in the prevalence of deepfakes in South Africa between 2022 and 2023, according to the recently published Sumsub identity fraud report for 2023, which was generated using internal data from the international verification company. This was the largest increase in the incidence of deepfakes recorded on the continent over the period. `Cheap fakes` still king Cayley Clifford, the deputy chief editor at Africa Check, told News24 that the fact-checking organisation still mostly dealt with what were called `cheap fakes`. A cheap fake is a piece of manipulative content that was created using relatively simple editing methods such as selectively clipping, slowing down, or speeding up a clip to remove important context or change the tone of the media. She said this type of manipulation is preferred as it is cheaper and easier to create. Clifford explained that it was currently quite expensive to make a deepfake in terms of the software and expertise that are required to make it convincing. She said that Africa Check had debunked some deepfakes, but there were not enough of these around to justify the use of advanced tools. `The tools that we currently use to debunk them are very much the same [that are used for debunking] regular misinformation. `We are not at a point yet where we are seeing them [deep fakes] enough to use them [advanced tools],` she said.