Amazon Removed Alexa's Privacy Setting. Your Meeting Recorder Never Had One.

Amazon just removed a privacy setting that millions of people had turned on.

Starting this month, Alexa+ processes everything in the cloud. The option to keep your voice recordings local is gone. If you had "Don't save recordings" enabled, too bad. Amazon says it needs cloud processing to power the new AI features.

The internet noticed. The story hit the front page of multiple subreddits. People were angry, mostly because it felt like a deal they had made was quietly taken back.

But the real story is bigger than Alexa. That setting only existed because Amazon got caught. After years of contractors listening to Alexa recordings, Congress asked hard questions, and Amazon gave users a way out. Now they've taken it back. "Privacy settings" on AI tools are not architecture. They're policy. Policy changes whenever the business model does.

Your meeting recorder probably doesn't even offer that setting.

What Alexa recordings and meeting audio have in common

When you say "Hey Alexa, set a timer," Amazon captures a few seconds of ambient audio. That's what people were worried about. That's what Congress held hearings over.

A recorded meeting is worse by an order of magnitude. A one-hour call with your team contains salary discussions, personnel decisions, unreleased product plans, client data your NDA covers, and candid opinions you'd never put in writing. Hours of your voice, not seconds. The people speaking didn't consent to being transcribed and stored on someone else's server.

Cloud meeting recorders hold all of that. On their infrastructure. Potentially forever, unless you go in and manually delete it.

The vendors aren't malicious. But the data is valuable. Voice recordings train speech models. Transcripts improve accuracy. Meeting content reveals what enterprise customers care about. There's a real incentive to keep it, and most terms of service quietly allow exactly that.

The opt-out problem

When you use a cloud transcription tool, you're typically agreeing that the vendor can use your content to "improve their service." The specific language varies. Some are vague. Some are specific about anonymization. Some give you a toggle to opt out of model training.

Amazon just showed you what that toggle is worth when the business model changes.

The only real opt-out is a system where there's nothing to send. Not a setting. Not a checkbox. A design where your audio stays on your machine because sending it anywhere was never part of the architecture.

How local transcription actually works

When MeetingVault transcribes a call, the audio never leaves your Mac. A local AI model, running entirely on your hardware, converts speech to text. Then the audio is discarded. There's no server receiving it. No database storing it. No vendor holding a copy they could hand over to a regulator, a hacker, or an acquirer.

The transcript stays on your machine. That's the only record.

It's not faster than cloud tools. Your computer does real work during transcription. But it's structurally private, not just policy-private. There's no setting for anyone to remove.

Why this matters more than it used to

AI is making cloud tools more capable. More companies are adding AI meeting summaries, action item extraction, searchable archives. All of that requires sending your audio somewhere. The more useful the AI features get, the more data moves to the cloud.

Amazon's rollback is a preview. As AI capabilities grow, the pressure to pull everything into the cloud grows with it. Vendors that built opt-outs under regulatory pressure will face business pressure to remove them. The ones that built local processing from the start don't have that problem.

If your meeting recorder sends audio to the cloud today, assume it always will.