Racial economic injustices of the past and present do not have to be our future. But that's what tools like potential output assume, so it's time for new tools.
I appreciate that you (and others such as Gbenga Ajilore) raise this point. I agree that it is very important to be aware of racial bias and that we should be extra careful when such bias is pointed out to us, because it may be so ingrained as to go unnoticed on casual inspection. However, at the moment I do not think that the issue you identified here constitutes or contributes to systematic racism. In short, the aggregate unemployment rate itself will always reflect the differences in unemployment rates between racial groups. The approach of disaggregating the estimate of the natural rate of unemployment at racial levels will only change the weights assigned to the different racial groups. Furthermore, precisely because of systemic racism, it is very useful to keep an eye on the disaggregated data.
I'll focus on the aspect of racism in the estimate of potential output. I have yet to grapple with your critique of potential output generally, but for this question we should take the potential output concept as given. The first question is how this actually works and compares to an approach without racial disaggregation. As I understand it, based on the CBO document you cite, first the natural rate of unemployment is estimated for each racial group separately (as their unemployment rate in 2005). Then an average weighted by population share is taken. So in 2005 the estimate for the natural rate is just the unemployment rate in that year. It would be the same if there was no racial disaggregation, where you would just take the overall unemployment rate in 2005. The difference between the two approaches comes in later years, when the weights of each racial group are updated, but not the estimate of of the natural rate for each racial group. So, if if the share of blacks and African Americans increased, while the share of whites decreased, greater weight would be put on the higher unemployment rates in 2005 of blacks and African Americans, leading to a higher estimate of the overall natural rate.
So, how large would such an effect be? I cannot find exact numbers on the change in racial composition of the labour force in the US between 2005 and now. But I can use some numbers for a back of the envelope calculation. On Wikipedia I find that the share of whites in the overall population declined by a 2.7% between 2000 and 2010. Let's suppose that 1) the rate of change was similar between 2005-2021, so 1.6 x 2.7%=4.32%. 2) To get an upper bound, let's suppose declining share corresponded to an increasing share for the black and African Americans only, who had an unemployment rate of 10% in 2005. Based on the graph whites had an unemployment rate of 5% in 2005. So the shift in weight would lead to an estimate of the natural rate of unemployment that is 0.0432 x 0.05= 0.22% higher than it would be in an approach without racial disaggregation. In general such a number will be quite small because we are multiplying two differences (change in population share and difference in employment rate) that are each in the order of a couple percent. (I do not know enough about macroeconomics to assess how large an effect a change of 0.22% in the estimate of the natural rate of unemployment would have on fiscal policy, but I would expect it is small.)
In general, the racial disaggregation approach doesn't seem racist to me, because it is only about adjusting the weights of the races. The systematic racism that causes differences in levels of employment will be present in the 2005 employment figures whether we disaggregate or not. It will also be present in the 2021 unemployment data. On the other hand, the adjustment in the population weights seems neutral to me. If the share of blacks and African Americans had increased, it would lead to a lower estimate of the natural rate of the natural rate of unemployment compared to the approach without racial disaggregation.
In fact, precisely because of the regrettable and undeniable presence of systematic racism, it seems to me to be actually useful to disaggregate unemployment data at the racial level. An important observation that you make is that the unemployment rate of blacks and African Americans was much lower in 2019 than in 2005 (from 10% to 6%). A possible interpretation of this number is that it corresponds to a decrease in the natural rate of unemployment for that racial group. This would have been hard to notice if we only focus on aggregate unemployment data. Even if we take 2019 as near the natural rate again, the difference in the aggregate between the unemployment rate in 2005 and 2019 would be small and perhaps not prompt a reassessment. It is only at the disaggregated level that we see that there is a need for updating the natural rate of unemployment for one racial group. (One hopes that this change corresponds to decreasing racism in the intervening years, but that is both speculative and very optimistic.)
The concept of "potential" was always a false and misguiding metric. When you factor in discrimination it makes the concept even worse. Hopefully it will be the reason to discard it from policy discussions once and for all.