Nothingburger. For the vast majority of users the audio was already getting sent to the cloud for realtime processing. Nothing here changes that.

For a few devices, some amount of processing was getting done on-device. Now, with the fancy LLMs, it has to revert to sending it all to the cloud.

Just turn on the feature to delete the recording after each processing. Best you can get if you want to keep using these devices and not run an NVidia processing cluster in your kitchen.

Not a formal audit, but a more recent review of the protocol: https://soatok.blog/2025/02/18/reviewing-the-cryptography-used-by-signal/