You don't generally need to accept licenses in order to scrape something, only if you want to distribute it.
The legal ambiguity comes from the question of whether LLM outputs are a derivative work of the training data. I expect that they aren't, but anything can happen.
