Note: Single-source report; awaiting corroboration.
A recent UN report indicates that nearly half of examined AI systems display gender bias, with many also showing race-related biases. This includes the association of women with home and childcare roles, and men with business and leadership, as found in analyses of 133 AI tools.
UN Women experts explain that large language models trained on historical data replicate societal inequalities, resulting in sexist and misogynistic outputs in about one out of five gender-related prompts. Some responses have described women as property or objects.
Experts emphasize that these biases arise from training AI on decades of unequal representation and highlight a significant policy gap due to limited gender inclusion in global AI governance frameworks.
Of 138 countries examined, only 24 have national AI strategies referencing gender, and 18 incorporate substantial gender-responsive initiatives. The report urges governments, companies, and developers to embed gender equality principles throughout AI system design and regulation.
The report also draws attention to increased online risks for women and girls, including AI-enabled abuse, and calls for comprehensive actions to mitigate these harms.