Protecting Biometric Privacy in Gaze Tracking Systems

This direction is split into other directions: biometric data sensitivity. Its last version stays readable and citable here.

Version 1 of 144 papers50 conceptsfrozen 2026-10-04

In the transition toward the Metaverse and ubiquitous spatial computing, eye tracking has emerged as the "secret sauce" for immersive experiences. It enables foveated rendering to save bandwidth, hands-free interaction for accessibility, and even diagnostic tools for healthcare. However, researchers like Süleyman Özdel and Eakta Jain warn that gaze data acts as a "biometric leaky pipe."

Unlike a password or even a fingerprint, your gaze is largely involuntary and continuous. It doesn't just show what you are looking at; it reveals who you are, how you feel, and even your future health risks. This has sparked a research direction focused on "privacy-preserving gaze tracking"—the art of stripping away a user's identity while keeping their eyes useful for interaction.

The Gaze as a Biometric Fingerprint

The core problem is that human eye movements are highly idiosyncratic. In Eye-tracked Virtual Reality: A Comprehensive Survey on Methods and Privacy Challenges, the authors explain that gaze patterns function as a high-dimensional biometric signature. Because these patterns are driven by unique neuromuscular pathways, they can inadvertently reveal a user's identity, gender, and even neurological health issues like autism or Parkinson’s.

This creates a "privacy paradox." To make a VR headset feel responsive, the system needs to know exactly where you are looking. But by knowing where you look, the system (or a malicious third party) can uniquely identify you across different sessions, effectively ending any hope of anonymity in virtual spaces.

Solving the Temporal Correlation Problem

Standard privacy techniques, like adding random noise to data, often fail when applied to eye tracking. This is because gaze data is not a series of independent points; it is a highly correlated temporal stream. If you simply add noise to each individual gaze point, a clever algorithm can "average out" the noise over time to recover the original signal.

To counter this, Onur Günlü and Enkelejda Kasneci developed a specialized differential privacy framework that specifically accounts for these temporal correlations. By mathematically modeling how one eye movement relates to the next, they can inject "smarter" noise that hides personal identifiers while maintaining enough utility for the system to recognize what activity the user is performing.

Sanitizing the Latent Space with Reinforcement Learning

Another intuitive approach moves away from adding noise to the raw coordinates and instead focuses on the "features" the computer sees. In Reinforcement Learning for the Privacy Preservation and Manipulation of Eye Tracking Data, Efe Bozkir and colleagues use an autoencoder to compress gaze data into a "latent representation"—a mathematical summary of the movement.

They then use a reinforcement learning agent to "sanitize" this summary. The agent is rewarded for two things:

  1. Keeping the data useful (e.g., the system still knows you are looking at a "buy" button).
  2. Making it impossible for a secondary classifier to guess your gender or identity.

By manipulating this hidden bottleneck, the system learns to strip away the "biometric signature" while leaving the "intent" intact.

The Risk of Implicit Gaze

The challenge becomes even more complex when we move from "explicit" gaze (using your eyes as a mouse) to "implicit" gaze (the natural way your eyes move while you think). Naveen Sendhilnathan and the team at Meta’s Reality Labs argue in Implicit gaze research for XR systems that these subconscious movements are the future of intuitive interfaces.

However, because these movements are subconscious, users cannot "mask" them. Mayar Elfares demonstrated in Gaze3P that gaze patterns can even reveal how private a user thinks an image is. If a user lingers on a sensitive part of a photo, their eyes are essentially voting on the privacy level of that content, providing a new vector for data harvesting that the user might not even realize they are providing.

Auditing the Immersive "Wild West"

Despite these technical advances, the actual deployment of privacy protections is lagging. A large-scale audit led by Chuan Yan in A Multi-Store Privacy Measurement of Virtual Reality App Ecosystem found that nearly a third of VR apps fail to disclose their collection of sensitive biometric data.

The research direction today is moving toward standardized protocols—like the "incognito mode" for VR proposed by Vivek Nair and Dawn Song—to ensure that as our devices get better at reading our minds through our eyes, they also get better at protecting the secrets they find there.

Papers in this version

History & citationVersion 1 · 2026-10-04

Split into other directions · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_protecting_biometric_privacy_in_gaze_tra_730fc205,
  title = {Protecting Biometric Privacy in Gaze Tracking Systems},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/protecting-biometric-privacy-in-gaze-tracking-systems/rdr_730fc2054d704845a1f5259373fd19aa},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-21},
}

This exact version (version 1, frozen text):

@misc{lacuna_protecting_biometric_privacy_in_gaze_tra_730fc205_v1,
  title = {Protecting Biometric Privacy in Gaze Tracking Systems},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/protecting-biometric-privacy-in-gaze-tracking-systems/rdr_730fc2054d704845a1f5259373fd19aa/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-21},
}