Yes that's giving me a bit of pause. I wonder what, say, Apple's ToS has for this kind of functionality.

Apple (and Google, and most others) process wake words on device. Imagine the technical complexity and expense of a billion devices streaming audio that is like 0.01% wake words.

> 8. I understand that LGE will retain the collected Voice Information for 6 months to fulfil the purposes for which we collected it, and after this period, it will be deleted or anonymized, depending on the case.

Somehow feels like it might be used for more than just wake word detection.

It's normal for home assistants to save what you said after the wake word.

Keep all the processing on device? Do so in a way that is legal in the jurisdiction you are selling the devices in without claiming it's the owner's responsibility to get consent from all individuals (even if they cannot legally consent because they are, say, a child!?) to comply with wiretapping laws in their own home? Choose not to sell an object that comes with software made to spy on you and sell all it learns to advertisers?

That fact that you read something like this as acceptable "legal boilerplate" just shows how, very sadly in my opinion, invasions of privacy have been normalized by these corporations.

In a place where recording audio requires consent from all parties, I don't think it matters whether "all the processing is on-device" given that the law was probably written with tape recorders in mind. If I buy a device that can record audio, and place it in my home where unsuspecting guests are recorded by it, whether that recording is transmitted to the manufacturer or stays within my home seems irrelevant.