Take the NYPD's use of proprietary "predictive policing" algorithms to identify where a crime may occur and identify individuals who may be a perpetrator.
These algorithms have long been feared to express the biases of their input data. [1]
The NYPD has obstructed freedom of information requests at every turn. They wouldn't even release their correspondence with the vendor because it "would reveal trade secret information." [2]
If we can't examine it or even know how it's being used, it's significantly more difficult to scrutinize.
[1]: https://www.propublica.org/article/machine-bias-risk-assessm...
[2]: https://www.brennancenter.org/our-work/analysis-opinion/cour...