There's no reason that a TV needs to be doing automatic content recognition on various inputs, but they're all doing it these days. Trying to 'de-evil' this sort of system is, first and foremost, fiddling around the edges of what's possible (I expect various people are reading and thinking, 'Oh, you think spoofing GPS will matter, cute!), but it's also remaining in the ecosystem that has, repeatedly, demonstrated that they're going to get their paws on everything they think they can justify, and then expand that over time. I assume the state of what's actually being done is far worse than what's in the papers, because someone, somewhere, though they could get a signal out of something. I've seen some fun papers of 'Well, you could do this awful thing.' (comparison of accelerometer data to deconflict which nearby phones are in the same vehicle vs separate ones to better refine social graphs), in addition to all the stuff we know is being done (ultrasonic signals in various ads, tracking shoppers by their wifi/bt beacon MACs, etc).
At this point, the 'sneaky snacky smartphone' approach to data collection (in which everything that can be collected is being collected, and probably used for things you can't imagine it would be useful for) starts to press heavily on the 'And I therefore shouldn't carry a smartphone' side of the scales.
