New theory:
Background: on my morning daily commute I have a bunch of different options. I can take the highway all the way to my work or get off on a bunch of different exists and take city streets. Because my work has a few paths of entry, some of which are closed at certain times during the day, I can't wholly rely on the car's navigation and Google's traffic reports to get me to the right entrance at the right time (since it has no idea about the various gate hours).
So very often the car will want to get off the highway early and drive on city streets into my work but I know that's a bad idea, because the gate it wants to drive towards will be closed, so I force the car into the left (passing) lane and turn off NoA while keeping AP engaged (via on the on-screen NoA button), so that the car definitely won't take the exit off the highway. And when I do this, 100% of the time the car WILL SLOW DOWN, thinking that it's missing its exit, and wanting to somehow correct the error. Since I know this is standard NoA/AP behavior (and definitely NOT FSD behavior, 2101Guy!--totally different thing), I am ready to lay on the accelerator or sometimes I just disengage AP altogether, depending on my mood.
So, with that background, here's the theory:
We all know that sometimes when you enter into tunnels, the GPS unit can get confused and put you onto a totally different part of the map than you're actually on--standard GPS unit errors. Maybe the driver had NoA on and upon entering the tunnel, the GPS unit got confused about where it was and felt that, in order to keep on the right path, it needed to take an immedate exit on the right. But then it believed it was going to miss that exit, so it felt the need to slow down, similar to my car on the highway when it's missing an exit. And thus why we see it slow AND try to change langes at the same time.