The ArcGIS 10.8 Extract by Mask help reads:
To process a selection of bands from a multi-band raster, first create a new raster dataset composed of those particular bands with the Composite Bands tool, and use the result as the Input Raster (in_raster in Python).
While no longer shown in the 10.8 help if you look at 10.3 help for Extract by Mask that reads;
If the input is a layer created from a multiband raster with more than three bands, the extraction operation will only consider the bands that were loaded (symbolized) by the layer. As a result, the output multiband raster can only have three bands, corresponding to those used in the display of the input layer.
Did you do the extract by mask outside of your ArcGIS session? Using Catalog or a model builder tool to do the extract from files on disk? If so, you ended up with RGB in the final output.
If you use Extract by Mask in ArcGIS and pointed the tool to the raster (with your chosen bands) in the table of contents then you ought to have the same bands in the output as the input. Check the symbology properties of the input data and compare those to the output data symbologies. Using percent clip as the default stretch value messes up the display in ArcGIS. Set both to none and report back.