For Heads-of · Practitioner

Proxy discrimination via correlated features

A model achieves discriminatory outcomes indirectly, through features that correlate with a protected characteristic even when that characteristic is excluded from the input.

  • high
  • fairness
  • bias
  • proxy-discrimination

How it happens

A protected characteristic is deliberately excluded from a model's input, but other features, postcode, purchase history, device type, correlate strongly enough with it that the model reproduces the same discriminatory pattern indirectly.

Why it matters

Removing the protected characteristic from the input feels like a fix but often isn't one; it can even make the discrimination harder to detect because the obvious check, does the model use this field, now returns no.

Mitigating controls

The controls that address this risk, ranked by effectiveness.

Framework and clause references

FrameworkClauseTitle
NIST Generative AI Profile (NIST AI 600-1)Harmful BiasHarmful Bias or Homogenization
UK General Data Protection RegulationArticle 5Principles relating to processing of personal data

Related resources

The external sources behind this risk, from the Resources library.