If they use audio-based ACR, they do know what's showing without needing to send any screen content telemetry.