Interesting thought. Probably possible, I think the main question there would be reliability for sensitive matters as well as security. Reliability meaning, if the AI is even a little off, it could cause a miscommunication with terrible consequences, and the best AI is probably still far off from comparing to an expert human interpreter (the best I know of that's fast and public is DeepL - similar to GoogleTranslate but arguably somewhat better - and then there are LLMs (Large Language Models) who can sort of do translation, but it's more of a gimmick than something they are designed for). There may be better though that's specifically in the sphere of Russian and Chinese language translation (I have no familiarity with AI translation tools originating from there). And security meaning, you'd need to be able to process what's said locally in such a way that it's not being sent off somewhere where it can be intercepted. For it to be local processing, it would require more local compute, which is going to be more expensive; might not be noteworthy difference between local and cloud compute if it's something like DeepL, but if it's a design more like an LLM, those can be greedy on GPU power.

So overall, I could see it being used as an assistive tool along with human translators to speed up the translation of especially long communications (if long communications is a thing in that context), but I doubt it's going to be replacing them meaningfully without worsening the communication process.