A new paper on arXiv asks whether generative AI contributes to 'beauty bias'—the societal tendency to reward people judged attractive by standards that are intertwined with gender and race. The researchers audited 27 text-to-image generators to see if and how these models may transmit such bias.

The abstract notes that societies reward attractive people, but it does not detail the audit's methodology or results. As a preprint, the paper has not yet undergone formal peer review, and the full findings are not available from the abstract alone.

This is a single source, so the article is a digest of the paper's stated aims rather than a synthesis of multiple studies. Readers interested in the outcome of the audit will need to consult the full manuscript.