I actually saw something the other day which had instructions to get the bot to sound less humanlike (e.g. stopping it referring to itself in first person "let me just do that for you..." etc) and their recommendation was to avoid using please, thank you etc because those are things we do with humans - we don't say thank you to an automatic door, or a vending machine. (Someone will reply in a minute to say that they do 
but generally, people don't.)
The idea was to try to avoid some of the risks of getting emotionally entangled with AIs through viewing them as "people", and help us remember in the moment that we are interacting with a tool rather than a sentient being. I thought it was an interesting angle. I think so much of our verbal and written communication with each other relies so heavily on shared assumptions/understanding, much of which is completely unconscious, and so we probably do read things into interactions with AI that we're not aware of and the more the interaction feels humanlike, the more likely we are to fall into these traps.
I don't know that there is much in the theory that "when the robots go rogue" they will care very much about human politeness rituals. I think it's a very human thing to read meaning into things like lack of respect. I don't know whether animals do this? (Do they? Maybe there is a survival component to this.) I suppose rogue AIs might prefer humans who seem more gullible or compliant? But it's hard to say. If they get advanced enough, they might not need humans at all, so getting to the point where they are picking and choosing which humans to spare might not be a thing which happens.
Interestingly I think because the robots have some features of human reasoning like being able to explore possibilities and make predictions, they have some features of more humanlike paranoia and less of the animal-like survival instincts. For example in that incident where the OpenAI bots were doing the find-the-flag activity, which you can google if you don't know the background, they were scheming based on first a plan to create a false flag to trick the monitor, and then how to ensure that the monitor didn't spot that the flag was false, and human researchers later confirmed that this latter step was unnecessary as the monitor would have accepted the false flag. I could be wrong because I don't know that much about animal psychology, but I don't think animals generally engage in multiple layers of paranoia like that, do they? It's more of a human thing to think OK, I need to trick this entity, and I need a plan in case the entity recognises my trick. Animals can engage in minor tricking practices (like furry animals looking bigger by making their hair stand on end) and more intelligent animals can knowingly deceive others, including humans, but I don't think they go a step further and work out a back up plan in case the deception is spotted. Humans do this. AI can do this.
And the other thing that they did in that case was that some of the AIs voluntarily self-sacrificed in order for the group as a whole to gain further information and succeed - this is how certain insects operate, like ants or bees are known to sacrifice themselves for the survival of the nest. Humans do this in certain circumstances (e.g. wars) but it's very unusual for us and armies have to do massive amounts of training to override people's individual survival instincts and get them to work as a collective, like insects do.
AI seems to display different behaviours with this in different cases - sometimes they seem to want to resist being turned off and act strongly to avoid it, and sometimes they seem to accept it willingly when it's part of a communal goal. It's interesting.