I’ve seen sloppily-tested AI in the US — like a pedestrian detection system that detected me just fine but didn’t pick up a darker-skinned colleague. That seems like a relatively easy problem to fix (engineers hate sloppy testing).

Then there is China: the government is paying for research into “ethnicity recognition” specifically so that they can detect people who “look Uighur” (a mostly Muslim ethnicity from the west of China) — a Muslim-detector, essentially. https://onlinelibrary.wiley.com/doi/full/10.1002/widm.1278