← Back to facegate.ai

Last tested: Pending

Demographic Fairness Report

Report version 1.0.0 — First test run pending

About This Report

FaceGate tests its face authentication system against demographic benchmark datasets to measure accuracy across race, gender, and age groups. This report publishes those results so tenants, regulators, and end users can evaluate whether the system performs equitably across the populations it serves.

Biometric systems that perform unevenly across demographic groups can cause real harm: higher false non-match rates lock legitimate users out; higher false match rates create security gaps. FaceGate treats fairness as a security property, not a compliance checkbox.

Test Methodology

Fairness testing uses the Labeled Faces in the Wild (LFW) dataset, a standard benchmark for face recognition systems. The dataset is partitioned by demographic group and the system is evaluated independently on each partition.

Primary metrics

  • FMR — False Match Rate: probability that two images of different people are incorrectly matched
  • FNMR — False Non-Match Rate: probability that two images of the same person are incorrectly rejected
  • Liveness accuracy — rate at which genuine live users pass the liveness challenge

Each metric is calculated independently for each demographic group. Results are then compared across groups to compute the differential ratio.

Fairness Threshold

FaceGate applies a maximum 3x differential threshold between the best and worst performing demographic group on any measured metric. If any metric exceeds this threshold, the system is considered to have failed the fairness test and the underlying model is not deployed until the gap is resolved.

MetricMeasured ByFairness Threshold
False Match Rate (FMR)Per demographic groupMax 3x differential
False Non-Match Rate (FNMR)Per demographic groupMax 3x differential
Liveness Detection AccuracyPer demographic groupMax 3x differential

Test Results

First fairness test pending

FaceGate has not yet completed its first formal fairness test run. Results will be published here after the initial test is conducted against the LFW dataset. The test will measure FMR, FNMR, and liveness detection accuracy broken down by race, gender, and age group, and will evaluate whether each metric meets the 3x differential threshold.

Fairness tests are run before each major model update and results are published within 30 days of test completion.

Last tested: Pending

Demographic Groups Measured

Results are broken down by the demographic attributes available in the LFW benchmark:

  • Race / ethnicity — as labeled in the benchmark dataset
  • Gender — as labeled in the benchmark dataset
  • Age group — divided into ranges (under 30, 30–50, over 50)

FaceGate does not collect or store demographic information about end users. Fairness testing uses public benchmark datasets only. No enrolled user data is used for fairness evaluation.

Related Policies

For information about what biometric data FaceGate collects, how long it is retained, and how to request deletion, see the Biometric Data Retention and Destruction Policy.

Questions

For questions about fairness testing methodology or results: