Generative AI tools like DALL-E and Stable Diffusion overwhelmingly default to depicting white men when generating images of professionals. A study found 97% of CEO images from DALL-E 2 showed white men, despite no user specification. This reflects how training data embeds societal biases into algorithmic assumptions, making unexamined patterns the user's default reality.

These aren't isolated bugs but systematic failures. Stable Diffusion rendered 100% of 'software developers' as male and 99% light-skinned, while real U.S. data shows 20% female and about 50% white. Medical profession imagery shows similar stark gaps, with surgeons depicted as 98% white men. The tools amplify stereotypes, consistently generating light-skinned 'attractive' people and darker-skinned 'poor' people.

The core dilemma for developers and designers is whether to ship products that force identity specification. This would add friction but make representation choices visible, challenging the myth of algorithmic neutrality. Major studies from UNESCO and Bloomberg confirm these patterns aren't edge cases but baked-in features of systems trained on biased data, raising urgent questions about equitable AI design.