Generative AI and Voice Cloning: Your Questions. Our Answers.
Even as artificial intelligence (AI) becomes more mainstream, credit unions have questions, as they should. As the industry learns more about it, AI has generated a great deal of both excitement and concern. Here, we explore the top questions that we at Curql Collective and our fintech portfolio company Illuma get about AI, along with our answers.
Q: How Easy Is It to Clone a Voice Using AI?
A: While today’s generative AI tools are useful to mimic a short, predefined clip of an individual’s voice, it is currently very difficult for AI to create accurate voice clones in real-time that can mimic natural and dynamic conversation.
Recently, there has been a lot of buzz about instances when financial accounts were hacked using AI voice clones. However, in most cases this was accomplished as a challenge undertaken by journalists publishing articles about concerns around voice clones. In these situations, the “hacker” could provide unlimited voice samples to feed the generative AI model and create only a short clone for a voice password. This is quite different from the process a bad actor would be able to replicate in a real-life account takeover (ATO) attempt.
Creating high quality voice clones using today’s generative AI requires a large quantity of high-quality voice samples. The best clones require superior AI technology and technical mastery, putting it out of reach for low-tech opportunistic hackers. Meanwhile, the most readily available generative AI software today doesn’t do a great job at replicating voices exactly. The resulting clones might be good enough to fool the human ear, but it isn’t designed to defeat voice biometrics security protocols at scale.
Our partners at Illuma offer voice verification systems to minimize vulnerabilities for financial institutions. As they tested this voice cloning process using multiple commonly available generative AI tools against their highly sophisticated voice biometrics technology, even their own experiments have resulted in voice clones that were blocked by their verification system over 95% of the time. All of this is good news for credit unions concerned about the threat of voice cloning.
Q: Is Biometric Voice Authentication Really Secure in the Era of Deep Fakes?
A: Illuma tells us they have a very high level of confidence in the security of properly designed and calibrated “passive voice authentication.” In contrast to “active voice authentication,” which only matches voiceprints for known and predetermined phrases, such as “my voice in my password,” passive voice authentication compares voiceprints based on the unique characteristics of a person’s voice in natural conversation.
Today’s voice cloning tools are not effective for replicating voices in real-time, making them unfit to use in dynamic conversations with another human. In fact, if a criminal were using these tools within a conversation with a credit union service representative, the lag time it would take to respond to questions would easily alert the credit union employee to the attempted fraud.
Plus, as we previously pointed out, high quality voice clones require a great deal of high quality voice samples, which makes the prospect incredibly difficult for criminals.
Q: How Can I Protect My Credit Union Call Center from Deep Fake Attacks?
A: In the face of increasing fraud attempts and the advancement of technology criminals use, it is important to have a multi-pronged security approach. Security questions alone are insufficient. With finely tuned technology for voice matching, contact centers and interactive voice response systems can better combat AI-generated voice clones and other synthetic voice-based fraud attempts.
In the end, credit unions should commit to always stay ahead of criminals when it comes to AI threats. With the right state-of-the-art fraud prevention tools, credit unions can better protect both themselves and their members from evolving threats. Visit www.Illuma.cx to learn more about how Illuma can delight and protect your members with seamless voice verification.