The prevailing theory is that they do the latter; the former would be both easy to detect (at the time of splice) and locate (via TDR).
Getting the signal out of the fiber pales in comparison, though, with the task of getting all of that data back to Maryland/Utah.
Unless they have specific cooperation of the cable owners and can tap/split the fibers at the landings, they must be spending a significant percentage of the cost of the original fiber runs (in parallel cables to return the tapped data). The mind reels.