There are two problems in play in this EXPO/XMP and SoC voltage disaster:
1) The common case of "Auto" voltages changing in automatic (pfft) way when EXPO/XMP is enabled, and they shove insane SoC voltages to make sure to minimize motherboard returns due to RAM that is not working at XMP. There are tiny variations here where if EXPO asks for 1.25V of DRAM voltage they will set like 1.27V etc. Nothing new really/
Do note that nothing in "settings changed" will show that voltage was adjusted -> it was AUTO and continues to be AUTO after enabling EXPO or XMP or overclock or changing some other voltage. You are at mercy of MB vendor to apply sane voltage while achieving their obviuos goal to avoid RMA due to incompatible RAM.
2) The extra retard version of this problem in Gigabyte BIOS. They have disaster level of BIOS support for AM5, two places to enter same values etc. So when changing SoC voltage, it might not apply or apply not the value you expect it to apply. Again "settings changed" would not indicate anything sinister, you really need to check HwInfo64 to see how voltages are.
It's not limited to voltages really, there are other interactions with memory timings and variuos settings that are in two places and have unpredictable interplay between them. Except other, non voltage related places result in frustated user ( remember that there is NO functional safe boot in this year 2022 platform from AMD, so each failed memory training => battery reset ). So no big deal for AMD, only enthusiasts suffer and even there a lower rung of them will just apply Buildzoid's timings for their memory IC vendor and not push things anywhere.