In a sign we're already beginning to live in a far-flung high-tech dystopian future where nobody can tell what's real anymore, a $243,000 heist has been pulled off by thieves using deepfake audio technology.
The Wall Street Journal reports that the first-ever crime using a voice generated by artificial intelligence (AI) has already taken place. According to the report, an unknown scammer managed to convince a CEO of an unnamed UK energy company that he was on the phone with the chief executive of their parent company in Germany. The voice asked that the CEO transfer €220,000 ($243,000) into the bank account of a supplier in Hungary immediately in order to avoid "late-payment fines", which the CEO proceeded to do.
Whilst on the phone, the scammers sent over financial details via email. Of course, the transfer would go to the scammers.
The scam took place in March this year, with the company's insurer Euler Hermes providing new details this week.
"The software was able to imitate the voice," Euler Hermes spokesperson Antje Wolters told the Washington Post. "And not only the voice: the tonality, the punctuation, the German accent."
Deepfakes of video as well as audio have become more and more convincing over the last year or so as this viral fake of Joe Rogan demonstrates. In another example, late last month a website for generating astonishingly convincing deepfakes of Jordan Peterson's voice was shut down after Peterson alluded to potential legal action.
The scam was shortly uncovered after the thieves got greedy, and called back the CEO demanding a further payment be made.
"'Johannes' was demanding to speak to me whilst I was still on the phone to the real Johannes!” the CEO said in an email seen by the Washington Post.
The CEO had grown suspicious when the scammers demanded the second payment and decided to call the CEO of their parent company directly, who picked up the phone.
The money that had already been sent has not been recovered, having been rerouted through accounts in Hungary and Mexico before being spread around elsewhere, and no suspects have been named or arrests made.
Before you go full Luddite and demand that the tech be destroyed, in addition to nefarious uses, this kind of tech designed to mimic natural voice patterns could have some really cool applications.
In a demo last year, for instance, Google's AI assistant was able to call a restaurant and a hairdresser and book a table and an appointment, without the human on the other end of the line ever knowing they were speaking to an algorithm.