It's not so much a loophole as they'd reasonably have expected mods that chopped up voice lines to make new sentences, and mods have been doing that for Bethesda games for years, sometimes with surprisingly effective results, but that's obviously super time-consuming and not as good as someone just reading aloud, let alone actually acting. Generative AI can suddenly chop up voice lines to make newer ones way faster with next to no effort, and give comparable quality to the original voice actor reading lines aloud, even though it can't do the acting part.

It's no skin off Bethesda's nose if people use generative AI to voice modded dialogue, but it could be a problem for the voice actors. Wes Johnson's done voice work for mods before, so mods aren't operating in a completely separate space to the voice actors who worked on the games. From a quick search, it doesn't seem like he charged for any of the work he did for mods (one was specifically for a charity fundraiser), but it wouldn't be immoral of him, or any other voice actor, to take paid commissions for mod dialogue. That's not as viable if generative AI can compete.

Anyway, none of this is really relevant to OpenMW specifically - sound files are game content, and we don't deal with game content