If face recognition makes the old racist "they all look the same to me" declaration, then peaceful protestors get arrested for looking like criminals.
“There are two ways that this technology can hurt people,” says Raji who worked with Buolamwini and Gebru on Gender Shades. “One way is by not working: by virtue of having higher error rates for people of color, it puts them at greater risk. The second situation is when it does work—where you have the perfect facial recognition system, but it’s easily weaponized against communities to harass them. It’s a separate and connected conversation.” [0]
[0] https://www.technologyreview.com/2020/06/12/1003482/amazon-s...
I mean "Our computer says we have footage of you robbing a supermarket" (higher error rates)
Another way for issues to arise (your easily weaponised point, above) is if you need a separate system for recognising people from certain groups. If it's all the same system, it's harder to argue that you are acting in good faith when you only follow up on matches on marginalised groups.