Well the device itself isn't powerful enough to run the speech recognition so they have to send the data back to their servers to run it (Nuance I think?) befor

Well the device itself isn't powerful enough to run the speech recognition so they have to send the data back to their servers to run it (Nuance I think?) before responding with what the user said so Siri can use it. It's certainly possible that Apple store every sound bite sent to those servers which is rather worrying. There needs to be some clarification on what they actually do with that data...