I had only a couple of clear nights to try it first with a Wratten #29 (deep red to IR pass) and only one night with a cheap Antlia brand 685nm IR pass filter.
Firstly, the good news is that the ASI462 appears to be sensitive enough to do IR-only guiding with a 55mm Borg objective (250mm focal length, Fluorite elements). Plenty of usable stars with 30 to 40 dB of gain.
The stars' FWHM "feels" a bit more bloated (could not focus to below 3), but that could be caused by seeing during my limited testing. The Borg 55FL is most likely to also be optimized just for the visual spectrum, rather than for the IR region.
But the stars appear quite round. So the bloating is likely caused by different parts of the spectrum having a different focal plane.
However, the autoguiding accuracy did not appear to feel significantly more stable than in the past when I was not using an IR pass filter for the 0.5 second exposures that my RainbowAstro RST-135 mount requires (at two regions of the 430 second period of the error curve of the Harmonic Drive gears, the first derivative of the curve is too large for even a 1 second lag time to keep up).
Again, that is with my very limited testing. Up in the Pacific Northwest I don't have enough clear nights in the Fall/Winter season to "waste" on autoguiding tests (swapping the filter in and out in the same night). I may need to dedicate a couple of nights, perhaps mounting the ASI462 on the main OTA (like my WhiteCat51 which has the ZWO EAF so I can at least focus from indoors). And perhaps mounting the IR Pass filter in a filter tray for fast swapping.
Since there was not any definitive positive or negative outcome so far, I decided not to post anything yet; until you prodded me :-) :-).
I still feel that multi-star guiding is the real solution to "seeing," like what is implemented in the Lacerta MGEN-3 and the INDIGO DONUTS. DONUTS in INDIGO (the open source is part of astroguider on github) is adapted from the DONUTS algorithm that was published in a professional astronomy journal). The DONUTS algorithm unfortunately requires good SNR (multiple bright guide stars), but since ZWO is already doing multi-star HFD in the Detect Star function, there should also be enough computational power in the Raspberry Pi to compute the average centroid for multiple stars. Computational astronomy is easier than optical solutions nowadays :-)
Chen