Right, these seem like reasonable hypotheses. I see a LOT of “innovation” happening in this space, though. In the future, or maybe even the present, I think it would be trivial to use speech to text and store conversations as small text files. Let’s say anytime it hears a specific brand name, “Cheerios” or “Toyota”, it records the conversation in a text file and sends it to marketers for research. It’s really not unthinkable.
The recent Mozilla report confirms that cars are using your microphone to determine what song or podcast you’re listening to, and listening to your conversations, so it’s not as if this is paranoid conjecture. If there’s money in it, and no rules to stop them, I think it’s almost inevitable.
I think automatic content recognition works by capturing still frames at strategic moments, so it may not take as much data as we think. For example, studios apparently hide watermarks that identify shows and movies. Then you would just need to make the tv detect the watermarks, not store and send screenshots of the screen. Then it can send a tiny CVS file of when and for how long you watched the show. It wouldn’t even need to know the name of the show. The watermark could be an alphanumeric code, and so even new shows would be detectable.
May I ask why you updated the firmware? I keep seeing people say that they disconnect the tv from the internet except for firmware updates, but if it works fine on day one and it’s always off line, why upgrade the firmware at all?