As WarrenTerra notes, it’s very unclear as to whether the postings being discussed here were the final, journal-typeset versions (usually not allowed per terms, but often posted), post-peer-review revised but author-typeset or revised manuscript versions (usually not allowed per terms, but very widely posted and hard to notice), or pre-peer-review manuscripts (usually allowed per terms, but dangerous in terms of scientific accuracy and consistency). Elsevier only allows the last of those to be posted, except, oddly, for allowing revised versions only on arXiv.
It’s very possible that this is over a clear breach of terms. However, Elsevier also has a restriction on posting anything for “commercial purposes” or in “systematic” ways or ways that “substitute for journal-provided services.” Their definition of commercial purposes explicitly includes advertising on websites, and extends even to companies letting customers or clients have copies of journal articles: they give an example of a pharmaceutical company providing a paper to a physician.
The “systematic” / “organized” / “substitute” restriction, also, is extremely vague, and appears as though it could cover almost any site trying to provide organized or convenient access to preprints or any other copies of papers. It’s possible that Elsevier is going after academia.edu as a whole using this restriction. That would be quite a bit more unsettling than simply going after the posting of a final version of a paper, as it would confirm the impression I get from their terms that the point is to make posting seemingly allowed but incredibly inconvenient.
