On live-git (via gentoo/kde overlay live-git ebuilds) but the bug has been around for a few months now.  Currently on qt-5.12.2 but I believe it triggered on qt-5.11 as well.  This may or may not be a dup of bug #401168.  I can test proposed fixes within a few days.

This applies to plasma on X, with multiple monitors.  My monitor layout is like this (ascii-art works best with monospace fonts):

          1
      +--------+
     2| 2160p  |3
  5   | pri    |
 +----+--------+
6|    |7     4     -> 1080p video/secondary
 +----+
   8

If that doesn't display correctly, here's the setup from xrandr:

HDMI-A-0 connected primary 3840x2160+1920+0
DisplayPort-0 connected 1920x1080+0+2160

So the primary is 4K with the top-left corner at 1920,0, with the secondary FHD diagonally to the left and below, top-left corner at 0,2160.

Edges 1,3,6,8 are at the external edge of the bounding rectangle (aka external struts), and an autohide panel placed on them works just fine.

Edges 2,4,5,7 are internal to the bounding rectangle (aka internal struts), and an autohide panel placed on them doesn't work.  The panel stays visible and over top any windows in the same area.

On the broken edges, hovering the pointer over the panel and then exiting does trigger an autohide, but the panel immediately pops back out, above other windows, as if the pointer had hit the trigger area again, despite actually being nowhere near it.

Unfortunately, I want my panel on a broken internal-strut edge, #4, and it won't stay hidden there at all! =:^(

Fortunately, the clue that it works on external-strut edges, but breaks on internal-strut edges, should help pin down the bug, and the fact that I'm on live-git and can rebuild with test patches and report results within a few days should help as well. =:^)