Did you know that deepfake software can not only fool humans, but they can also outsmart voice authentication systems with alarming accuracy?
Yes, you read that right.
According to a 2023 global study, cybercriminals only need three seconds of audio to clone someone’s voice using artificial intelligence (AI) powered tools. To make matters worse, computer scientists at the University of Waterloo have recently discovered that deepfake software can fool voice authentication systems, a security measure used by many corporations and financial institutions, with up to a 99% success rate.
At a time when the rising popularity of generative AI and machine learning has raised concerns about online privacy and digital security, these research studies certainly paint a terrifying picture of a future where cybercriminals can access your bank account or cause severe emotional damage in a matter of seconds using easily available AI-powered tools.
While there is no denying that deepfake technology has the potential for creative and practical applications, such as in the entertainment industry or for voice-overs, it also has the potential to be misused by malicious actors. Deepfake voices can be used to create misleading audio recordings with the intent of spreading misinformation or impersonating an individual without their knowledge.
Needless to say, researchers outsmarting advanced voice authentication systems with deepfakes in less than 10 tries has rung the alarm in relevant industries. However, before we delve into the specifics of the experiment, let’s take a brief look at what is a deepfake.
What is a deepfake voice and how is it created?
A deepfake voice refers to using AI and machine learning techniques to manipulate or clone audio in a way that mimics a specific individual’s voice. It involves using algorithms and neural networks to analyze and learn from existing audio recordings of a person’s voice. It then generates new audio that sounds like them, even if they did not say those specific words or phrases. AI voice scams are famously used to dupe people into lending money to the scammer, pretending to be one of their loved ones.
As complex as it may sound, cybercriminals can easily create a deepfake voice by training AI models on a dataset of recordings from the target individual, capturing their speech patterns, intonations, and other vocal characteristics. Once trained, the model can generate new audio that closely matches the target’s voice and even create entire audio clips that appear to be spoken by the individual.
Read More: Beware of Deepfakes – How AI is being exploited by blackmailers to generate explicit content
Deepfake software vs voice authentication
Voice authentication systems offer a convenient and secure way to verify someone’s identity based on their unique voice characteristics. Banks, government agencies, and companies worldwide use this technology to prevent unauthorized access. However, a team of scientists at Waterloo University found a way to bypass this system within 6 tries with alarming accuracy.
“When enrolling in voice authentication, you are asked to repeat a certain phrase in your own voice. The system then extracts a unique vocal signature (voiceprint) from this provided phrase and stores it on a server,” explained Andre Kassis, the lead author of the research study. “For future authentication attempts, you are asked to repeat a different phrase, and the features extracted from it are compared to the voiceprint you have saved in the system to determine whether access should be granted.”
Hackers and scammers try to exploit voiceprint technology by using deepfake software to generate convincing copies of someone’s voice, which has prompted developers to introduce countermeasures to identify machine-generated speech. However, the researchers at Waterloo developed a method that evaded these countermeasures and essentially made the deepfake audio indistinguishable from the authentic one.
In tests conducted against voice authentication systems like Amazon Connect, the researchers achieved high success rates in fooling the security measure within a few attempts. The report argues that existing spoofing countermeasures are flawed and need improvement. While voice authentication provides additional security, further advancements are necessary to address the growing threat of deepfake voice scams.
How AI tools are changing the game for cybercriminals
The increased sophistication of AI tools has allowed cybercriminals to grow and evolve their methods, empowering malicious actors with capabilities that previously took years to master. Today, artificial intelligence tools can create highly convincing deepfake content, automate attacks at scale, and exploit vulnerabilities more effectively. The use of AI-driven techniques by cybercriminals also presents new challenges for cybersecurity professionals and encourages constant innovation to stay ahead of evolving threats.
Nevertheless, staying vigilant, developing critical thinking skills, and adopting responsible digital practices are crucial to protecting your confidential information and mitigating the risks associated with deepfake technology.
This brings us to the end of our blog on how deepfake software can trick voice authentication. If you want to learn more about how AI is changing the landscape of digital security, please head over to PureVPN Blog.
Moreover, you can also take a look at PureAI(alpha) – an advanced virtual assistant that allows you to explore the world on your terms without compromising your privacy. PureAI(alpha) is exclusively available in the Member Areas on PureVPN, so make sure to check it out.
Read More: Deepfake defense – How to support and protect yourself from AI manipulation
