Because that's what this leak seems to be implying.
Are you looking at the evidence or the articles written about it? Because the evidence is simply that there exists a method to select Tor-related traffic out of a stream. That alone is nothing. You know very little about what the input is or where the output goes, or what conclusions are drawn by the people analyzing it.
To reach the paranoid conclusions that are getting such traction around here one has to presume that the same institution competent enough to construct this collection apparatus is so fantastically braindead as to make, on a continuing basis, elementary mistakes of analysis on the data that would undermine both their avowed purpose and whatever totalitarian scheming one wishes to imagine is "really" going on. Mistakes which, somehow, miraculously, every single reader with the slightest bit of understanding of the subject matter detects and avoids! I don't buy it.