Practical Considerations Completely Overlooked

I took great interest in Sam Altman's recent comments about deepfakes during an interview at the Federal Reserve's Integrated Review of the Capital Framework for Large Banks Conference in Washington D.C. on July 22, 2025. Essentially, he stated that with just a few seconds of speech, powerful AI models can perfectly reproduce someone's voice -- and that it's "crazy" that some financial institutions are still using voice biometrics for authenticating transactions. His bottom line: "AI has fully defeated" voice biometric authentication. As someone with 19+ years of experience in the voice biometrics industry, I couldn't disagree more with his seemingly complete dismissal of voice biometric technology. In making this statement, Sam Altman is completely ignoring many practical, real world considerations.

Before moving on, there are many articles covering his interview. If you haven't seen or heard any of the news coverage on the topic, click here for a good summary from Fortune.

We Agree on One Thing - Deepfakes ARE a Concern

There's one thing I won't disagree with Sam Altman on -- deepfakes ARE a big concern to every company providing biometric authentication products and services, as AI-based attacks have been shown to be able to bypass biometric-based voice, facial recognition, fingerprint, and iris systems under the right circumstances (more on this below).

No one in our industry takes these possibilities lightly. As an example, our company has been actively researching and developing deepfake detection technology for over 6 years. We now spend nearly as much time researching and developing deepfake detection tools as we do on our core algorithms that support voice biometric speaker verification and identification. Deepfakes have become part of our everyday reality in recent years. And like anti-virus software companies -- we've developed effective tools to address deepfakes, which can be frequently updated in real-time.

The Threat of Deepfakes is NOT New

Deepfake attacks and news articles surrounding the threats posed by deepfakes are not new -- they've been around for several years. Over 2 years ago I wrote about the threat of deepfakes. I was concerned then, and I remained concerned now. Here's a link to the original article. Even at a couple years old, it remains relevant and worth reviewing.

The original article is quite lenghty and covers many considerations, so I won't repeat the details here. However, there are two key points to make here. First, concerns about deepfakes are not the same as panic, nor do they represent a need to abandon voice biometric technology. And second, it should be clear that Sam Altman has glossed over a number of very important practical considerations related to voice biometrics. Below I'll summarize some of these key considerations:

What Sam Altman is Missing

Sam Altman claims that AI-based models can circumvent biometric authentication systems. Yes, in certain circumstances, AI models can create outputs that will bypass voice biometric systems. My argument is to clarify this claim with "under the right circumstances". And these circumstances are for all of us to forget about the real world. Several key points he is glossing over:

• Voice Biometrics vs. No Voice Biometrics. Our business is growing each year. More and more companies are adding voice biometrics to their mobile applications, IVR systems, and call center agent software as security professionals continue to understand the importance of this factor. The simple fact is, speech-based systems employing voice biometric technology will always provide far greater protection compared to systems not using this technology.

• Multi-Factor Authentication (MFA). I've been involved with the deployment of voice biometric systems since 2006. Never once have we deployed a system where voice authentication is the only security measure. EVERYONE we work with uses multiple security factors. Even if we assume that voice biometric technology could be cirmcumvented by a deepfake -- there would still be multiple other factors that would have to be circumvented. The reality is, a well-implemented MFA system, utilizing voice biometrics as a factor, will remain extremely difficult to beat.

• Access and Acumen Sam Altman's message implies that everyone has ready access to this technology and the technical acumen to leverage it to defeat systems protected by MFA. This is simply not the case. Statistically speaking, there is a very low risk for widespread, automated deepfake attacks to occur within the large scale mobile app, IVR system, and call center operations we help to protect.

• Deepfake Detection EVERY reputable voice biometric software vendor that I know is also providing deepfake detection tools. We've been providing them for a couple years now and make them available to all of our voice biometric authentication customers. I won't say that it's impossible for a deepfake to pass detection with a voice biometric authentication system; rather, it's a matter of employing strong countermeasures (like all anti-fraud security approaches). We have such countermeasures and we're improving them constantly. We're far from the death sentence to voice biometric technology that Sam Altman infers.

Conclusion

Deepfake attacks against systems protected by voice biometric technology are a reality. However, after several years of operating commercial voice biometric systems with deepfake scanning technology, we're just not seeing any evidence of widespread attacks. Yes, there are occasional deepfakes that are presented to our systems, but statistically speaking the threat is relatively low -- and certainly does not warrant abandoning voice biometrics. And it's worth noting that on a daily basis, fraud and risk professionals face a far greater comparative threat from "friendly fraud" within systems that are not secured by voice biometrics.

I disagree with Sam Altman and his conclusions about voice biometrics. However, the future of authentication isn't about picking sides; it's about building layered, intelligent systems that stay ahead of threats. And simply put, voice biometric technology remains a powerful tool in that arsenal.