I am curious, when did the decision to make GPS not follow leap seconds occur? had you asked me before I read your comment, I would have said that it doesn't honor leap seconds and never has.
TAI is also, of course, fundamentally just a divergence from UTC caused by not applying leap seconds. However, formally, UTC is defined as TAI plus or minus the leap second offset in order to minimize divergence from UT1 (which is a smoothed astronomical time base rather than monotonic). GPS time is TAI plus or minus a different offset caused by some of the leap seconds being considered.
There is a basis for this, LORAN did the exact same thing but with an earlier divergence point.
I think all the online converters I found a couple of years ago failed to get this right.