Debunking Myths Part 1: Addressing Bias in Biometrics

Biometric technology has transformed the way we authenticate identity, delivering both convenience and enhanced security. However, like any transformative technology, it has faced scrutiny and generates concern —particularly around the issue of bias, legality, privacy, covert usage and the risks associated with AI. While the concerns are valid, they are often misunderstood or even exaggerated. 

This blog is part of a five-part series dedicated to unpacking and debunking common myths surrounding biometrics, offering clarity on where the challenges lie and how the industry has been addressing them.

We start with one of the most discussed topics: bias.

See the other posts in this series:

• Legal Frameworks
• Template Security
• Device Storage
• Universality

Myth 1: All Biometrics Are Inherently Biased

One of the most prevalent myths is that biometric systems are inherently biased against certain demographic groups, such as people with darker skin tones or women. While early iterations of some biometric technologies exhibited these issues, this is not an inherent flaw in biometrics as a whole. Rather, it reflects historical challenges in data collection and algorithm training.

Modern biometric systems are designed with diverse datasets that better represent the global population. Moreover, advancements in machine learning allow for continual refinement of algorithms, reducing disparities in accuracy across demographics. Responsible developers now prioritize inclusivity as a core design principle, ensuring fair and equitable system performance.

Myth 2: Bias in Biometrics Is Worse Than in Other Authentication Methods

Critics often argue that from a bias perspective, biometrics is a poor choice compared to other authentication methods, such as passwords or tokens. This perspective overlooks a critical point: all authentication methods have vulnerabilities and limitations. For example, passwords are prone to theft, guessing, and social engineering, which can disproportionately affect less tech-savvy users.

In contrast, biometrics offer a personalized and secure authentication method. They are the only way to link a person to their identity in a way that cannot be stolen or phished. The industry’s proactive approach to addressing bias, combined with ongoing improvements in accuracy, positions biometrics as a more inclusive and secure alternative to traditional methods.

Myth 3: There Is No Way to Measure Bias in Biometrics

Another misconception is that bias in biometric systems cannot be effectively measured, leaving developers and organizations without a clear path to improvement. In reality, extensive testing frameworks and evaluation methodologies exist to assess bias and ensure equitable performance across demographics.

One of the leading efforts in this space is the National Institute of Standards and Technology (NIST) Face Recognition Vendor Test (FRVT). NIST conducts rigorous, independent testing of facial recognition algorithms, analyzing their accuracy across different demographic groups. The results provide valuable insights into disparities in performance, helping developers identify and address potential biases in their systems. 

For example, recent NIST reports show that while earlier facial recognition systems had higher error rates for certain demographics, modern algorithms have achieved substantial improvements through better training data and enhanced algorithm design and there are notable differences between the algorithms in this regard. The following two charts demonstrate the top level performance achieved by the ROC algorithm to illustrate the point.

According to ROC’s Chief Scientist and Co-Founder Dr. Brendan Klare said, “Over the last decade, facial recognition research has delivered not only orders of magnitude improvement in overall accuracy but also balanced improvement across all demographic populations. In NIST testing, ROC and other market leaders deliver astonishingly low error rates across all demographics, which error rates across demographics are more or less effectively equal within the statistical bands of uncertainty. ROC and other elite peers continue to innovate at a whirlwind pace to improve accuracy both overall and for all demographic populations by applying cutting-edge machine learning scientific techniques.”

Myth 4: Biometrics Reveal Demographic Factors and Exacerbate Human Bias

A pervasive myth suggests that biometrics inherently reveal demographic factors such as gender, race, or ethnicity, and that this visibility exacerbates human biases. In reality, biometric systems, when properly designed and implemented, focus solely on unique biological patterns—such as fingerprints, facial geometry, or iris structures—for identity verification. These patterns do not need to be linked to or reveal demographic characteristics.

Unlike traditional identifiers, such as names, zip codes, or addresses, which can implicitly disclose a person’s background, biometric data can be processed in ways that avoid exposing sensitive demographic information. For example, privacy-preserving biometric frameworks store and match data in fragmented, encrypted formats using technologies like Multi-Party Computation (MPC) and Zero Knowledge Proofs (ZKP). This ensures that even if the system processes a biometric match, no demographic details or the original image or biometric sample are accessible or revealed.

Media reports often highlight incidents where biometric systems have been linked to discrimination against certain demographics, but these issues typically stem from flaws in deployment and operational models, not the technology itself. Many of these cases arise in the law enforcement domain, where facial recognition systems are sometimes misused as the sole basis for arrests. In reality, these systems are designed to return a subset of potential matches for investigators to review, serving as a tool to aid decision-making rather than replace it. Proper vetting and corroboration with other evidence are essential to establish probable cause. A facial match alone should never be the sole justification for an arrest, underscoring the importance of responsible deployment practices to avoid misuse and ensure fairness.

This point is underscored within the principles of the EU AI Act which is designed to ensure that real-time biometric identification systems are used as investigative aids rather than definitive decision-makers. By requiring transparency, accountability, and human oversight, the Act aligns with best practices that dictate a facial match should only provide a starting point for further investigation, not serve as standalone evidence for actions like arrests. Unfortunately, the United States does not currently have a comparable law so it is incumbent on solution providers and practitioners to incorporate best practices in their deployments.

Charting a Path Forward

Biometric technology is not perfect in the sense that there is no 100% performance, and there is always a tradeoff between false accepts and false rejects, but neither is it the biased, irredeemable system that critics sometimes portray. Myths surrounding bias often obscure the significant progress the industry has made—and continues to make—toward inclusivity and fairness. As innovation continues and biometrics continue to be adopted more and more, the conversation should focus on supporting the development and deployment of fair, accurate, and ethical systems. By debunking these myths and working off facts, we can create a future where biometrics are trusted, secure, and truly inclusive.