At scale, being confident your data is authentically pre-2024 is going to be hard unless you are maintaining the corpus already! Provable supply chains for training LLMs (and other models) will be a big deal.
PS - "every cousin from your aunt's side" - is this a common saying somewhere? Love it.