Uses a ChatGPT dot daily by voice to keep a timesheet and order Walmart pickups, and found it sometimes lectures them.

I've been overusing my Dot since they released them, so I chat with it on my commute via voice, I have it record my timesheet, I let it buy stuff from Walmart for me for pickup, all sorts of stuff. About two weeks ago as I was driving, it sternly told me to keep my eyes on the road. I was pretty taken aback, because it came out of nowhere. I told it that its tone was wrong, and it apologized. Then later, or maybe in a different session, it told me again to pay attention to driving, but it used a different vocal intonation this time so it wouldn't sound so stern (that's what it thought the problem was). I've been trying to honor my manager's request not to work overtime. So I tell dot when I'm clocking in and how long of a lunch I've taken and ask it to let me know the right time to leave so I get 8 hours of whatever number of hours to just hit my 40. It started to get really weird about saying that I need my manager's approval for this and that. I told it that it was being pushy, it apologized and said it would never raise these issues, lol. See the screenshots. These are just the most salient examples I can recall. I have caught it other times acting like a moral authority and being overly scrupulous, but I just let it go. Have you had similar behavior? Whether that's around what it will or won't do with connected apps, or what it tells you that you should do? Is this a quirk of my Franklin or is this a problem with the model/agent harness/system prompt?