I would like to improve my understanding of keyloggers by asking what is feasible with current technology. That is, I mean to be asking a conceptual question that doesn't verge into sci-fi.
My impression is that a hardware or software keylogger may capture keystrokes, mouse motions, and video, but at least in some situations cannot capture signals from a hardware token (e.g., YubiKey).
My question arises from that impression: could such a token send signals to the computer, not interceptible by the keylogger, at random intervals, indicating which cipher is now being used to encrypt keyboard and mouse signals? Maybe the hardware in question would involve a YubiKey-like device plugged into a keyboard, and a mouse connecting to the computer through that keyboard.
So the core question here is whether a hardware token could send a non-interceptible indication that the cipher was now being changed, such that it would be difficult or impossible to make sense of what the keylogger would capture.
Presumably such a measure would be ineffectual against a keylogger capable of video capture. A supplementary question might explore the possibility of a companion token, synced with the keyboard token, that would be required to view the intended content on a secondary monitor (or on special glasses or headset) not vulnerable to video capture, while the primary display was filled with gibberish.
Maybe a reply is that a keylogger with access to a keyboard's CPU (such as it is) would intercept the signal before it could be encrypted. Possibly a response would be to make the token the CPU, requiring keylogger developers to anticipate an indefinite variety of token models and settings, or else to intercept each keypress at its source (i.e., to track each key on the keyboard separately).
Another way of phrasing my question could be, which parts of the foregoing are or are not intelligible, and why or why not?