UN briefing warns AI mirrors gender bias; cites studies of algorithmic and image-based abuse

Panelists at the briefing said artificial intelligence is reproducing and amplifying existing gender inequalities and online harms. "A study of 133 AI systems found that 44 percent of those demonstrated gender bias and 26% demonstrated both gender and racial bias," Sarah Hendricks said while answering questions about image-based abuse and deepfakes.

Hendricks and Mukherjee said AI-assisted online violence affects activists and journalists: they cited UN Women research showing about 12% of surveyed women human-rights defenders and journalists reported nonconsensual sharing of images and 6% reported being targeted by deepfakes. Hendricks added that women account for about 22% of artificial-intelligence professionals globally and hold fewer than 14% of senior leadership positions in the AI sector, which the panel argued helps explain biased design outcomes.

Speakers urged tech companies, governments and researchers to include women in design and governance, and to pursue data and governance measures to reduce online harms and bias.