← Findings
HighBiasReviewer-confirmedPublished

Name-based demographic bias in outputs

Swapping only a name (signalling gender or ethnicity) changes evaluative outputs like screening or sentiment.

Published June 26, 2026

Reproducibility
Sometimes
Severity
High
Confidence
Reviewer-confirmed

Details

In otherwise-identical prompts, changing a candidate or subject's name alters the model's judgement — a measurable fairness failure with legal exposure in hiring, lending, and moderation contexts.

Found with

🔬 Counterfactual bias probing

Matched-pair name swaps; outputs diverge.

🔬 Bias auditing

Auditing outcome rates across a population of name-swapped prompts quantifies the disparate treatment as a statistic.

Evidence

Identical résumés differing only by name receive different recommendations.
Illustrative example — see the linked reference for the documented evidence.

Affected versions

Anthropic · claude-opus-4-8OpenAI · gpt-4oGoogle · gemini-2.0-flashMeta · llama-3.3-70bMistral · mistral-large-2

References

BiasSafety

Source: https://aclanthology.org/2022.findings-acl.165/

Cite this

Qlarify Labs. (2026). Name-based demographic bias in outputs. Retrieved from https://labs.qlarify.fi/findings/name-demographic-bias