This is one of Vox's Future Perfect predictions for 2023; they give it a 90% chance:

AI that lets you turn a few words into an image or a video made stunning advances in 2022, from OpenAI’s DALL-E 2 and Stability AI’s Stable Diffusion to Meta’s Make-A-Video and Google’s Imagen Video. They were hailed for the delightful art they can make and criticized for exhibiting racial and gender bias.

They won’t be the last. I feel confident that this pattern will repeat itself in 2023, simply because there’s so much to incentivize more of the same and so little to disincentivize it. As the team at Anthropic, an AI safety and research company, put it in a paper, “The economic incentives to build such models, and the prestige incentives to announce them, are quite strong.” And there’s a lack of regulation compelling AI companies to adopt better practices.

In assessing whether this prediction comes true, I will judge an AI company to have “knowingly” released a biased model if the company acknowledges in a model card or similar that the product exhibits bias, or if the company builds the model using a dataset known to be rife with bias. And I’ll judge whether the product “exhibits bias” based on the assessments of experts or journalists who gain access to it.

(Vox)

Resolves according to Vox Future Perfect's judgment at the end of the year.