IIRC yes this is a bug of some type when using it on hard edges.
I don't believe that is a bug, but a result of simple physics. Whenever you have a dark object against a light background, there is generally a degree of Chromatic Aberration created in the image due to the very slight differences in the way light of different wavelengths (Color) travel through the lens. This is the most likely cause of the halo when using the brush. You can try to remove this by first going to Lens Correction and checking the box to remove Chromatic Aberration. If that does not work, try your hand at manual correction. Results will vary. Generally it is best to avoid making major brush adjustments in this case.
Joe, your answer doesn't make much sense to me. I don't see what physics has to do with it. I do understand what causes chromatic aberration.
As a matter of course I always remove chromatic aberration and enable profile corrections. I don't believe either has any effect on edge detection. In this case there is none, just a clean edge which auto-mask won't go within a pixel of. That doesn't sound like desirable behaviour to me.
Avoiding major brush adjustments where there is a hard edge is certainly an option but not something I would be looking for in a premium photo processing package.
I would be interested in work-arounds other than tedious hand painting.
Chromatic Aberration is not always complete removed. Even if it were, physics is still involved. The "clean" edge you are referring to is not as clean as you think, except perhaps for vertical or horizontal edges. Pixels are square with vertical and horizontal edges. A diagonal or curved edge is not "clean" as there is some smoothing going on to add realism to curved and diagonal lines (Anti-Aliasing), which is part of the problem you are seeing; this also results in difficulties in removing Chromatic Aberration. Magnify your image to 10 to 1 to view individual pixels and you will see that the edge is not "Clean". Premium package or not, physics makes this a very difficult task. You are pushing the math beyond the capability of auto mask.