Face recognition might be usable to spot agent provocateurs at protests but can be a double edged sword

I'm extremely skeptical that facial recognition would be better at this task than a group of trained humans (although I'm personally terrible at faces 😆). Existing facial recognition software is trained on datasets that reinforce existing oppressive biases, so you'd probably have to train from scratch on a fresh dataset or do some extremely careful transfer learning to even begin to get anything useful. (IMO getting a sufficient dataset is the hard part since you can do basic facial recognition with an eigenvalue decomposition on any computer.)