Eh, think about this from the perspective of a potential scammer who doesn't speak the targeted language very well, or doesn't have someone adjacent to them who does. (ChatGPT can do many human languages)

Machine translation worked 'okay' for scammers in the past obviously, but it's no replacement for the 'natural language' that these models can produce, that is surely much more effective.

To me, that's a game changer.

To give a hypothetical, let's presume that in, say, Kyrgyzstan there's not very much successful spam today, because the spam is predominantly in English and not well targeted. In this environment, one could (today!) use chatGPT to generate realistic spam in Kyrgyz, and potentially see a high conversion rate as that's potentially not been done well before.