The worst thing is that there is nothing technical preventing this kind of service from running fully local. All the audio is already being routed to any participant and transcription models are lightweight enough and can run in virtually any computer.