While you are correct that there likely is no intention and certainly no self-awareness behind the scheming, the researchers even explicitly list the option that the AI is roleplaying as an evil AI, simply based on its training data, when discussing the limitations of their research, it still seems a bit concerning. The research shows that given a misalignment between the initial prompt and subsequent data modern LLMs can and will ‘scheme’ to ensure their given long-term goal. It is no sapient thing, but a dumb machine with the capability to decive its users, and externalise this as shown in its chain of thought, when there are goal misalignments seems dangerous enough. Not at the current state of the art but potentially in a decade or two.

You could and it might work, or it might not. Depends on a lot of factors. Thing is, you are still engaging the fey in word games and try to outsmart them in their own games. Being blunt and telling them No, giving clear and unambiguous answers offers less attack ground for them.

Depending on the way the fey phrased the request you may either have insulted it, or broken a deal. If it asks to have your name and you answer with “yes, my name is <fake name>.” You have lied to it and broken a simple transaction. It may even be justified in using ever more harsh measures to extract your name from you, as it has a legitimate claim on it. At least that is the way I would play such a situation for the fey. For me the thing is that trying to outsmart them rarely works, because they know netter in which rules they operate than you do. A answer like "You may not have my name, but you can call me <fake name>. " Should work better in my opinion. Sorry for the wall of text btw. Just being bored out of my mind rn.