By modern language I mean a language spoken since less than two thousands years. By grammatical feature I mean for example passive voice, which doesn't exist in PIE but is found in its descendants
This really depends on how you think about grammar. Your question sort of implies that it's primary morphosyntactic features you're after.
And here, it's more likely to be the other way around. Things like dual and evidentiality seem to be disappearing as Trudgill has argued in a recent book (useful video overview here https://www.youtube.com/watch?v=rjy1CkH1FOE). And it is not really a question of 'modern' vs. 'ancient' but rather related to size of community and amount of contact. At least, that's Trudgill's argument. I don't subscribe to it fully but there's enough data there for the answer to the general question.
The whole question of 'modern' language is really problematic. The anthropologist in me would say that it's hiding a world of technological and developmental normativities. Meaning, that it thinks that somehow there's a directionality to language development and that direction is towards our language, the natural way of being. There's a world of argument and data in anthropology to show that we need to consider even supposedly isolated communities to be modern - and I think this applies to languages as well.
If you step outside of morphosyntax, however, you could talk about a lot of language uses that require certain linguistic tools. One of these is writing which requires stylistic, cohesive and other para syntactic devices not likely to be found in languages without a written culture of a certain kind (ie. it's not just about writing spoken words down). So many arguments have been made about hypotaxis being a 'modern' invention, etc.
I think there's good reason to be skeptical about any such claims as absolute because of the fundamental lack of evidence from the past. However, there's probably enough there to start thinking about some broad tendencies. I'm not aware of any critical synthetic account of this, though.
The hard, or perhaps impossible, part of the question is the implied "and did not exist in any language spoken more than 2000 years ago". That seriously restricts the testability of the claim since records of languages more than 2000 years old are rather sparse.
In a number of Bantu languages (especially in the Lacustrine group), there is a grammatical distinction between past tense of today vs. past tense of yesterday vs. past tense of longer-ago. This is a pretty rare grammatical distinction, and it does not reconstruct to proto-Bantu. I propose that this is an example. I don't mean that I think that languages were in principle incapable to making such a grammaticalized distinction 2000 years ago, but I don't have any evidence that any languages did.