Do Whisper-based apps send your audio to OpenAI?

Some do and some do not. Many apps built on Whisper run the model on your device. Others send your recording to OpenAI's API, billed to a key you enter, and a few send it to a Whisper server you host. The name on the app tells you nothing about the path, so check the settings and the documentation, and then test the app yourself.

Three paths, and only one keeps the audio on your own device

Whisper To Input, an Android keyboard, offers the OpenAI API with your own key or a Whisper server you host yourself ¹. Both routes are legitimate. Only the second keeps audio on infrastructure you run, and even then the phone must reach your server.

Superwhisper is the clearest example of an app with both paths. On its Free plan, dictation uses local Whisper models without AI processing, and the company says nothing dictated leaves the device ². Its Pro plan adds cloud models and AI processing ³. When you use your own key, the audio goes to that provider under your account, so the provider's data terms apply as well as the app's. Anyone who wants a bring your own key setup should read those terms, not only the app's.

MacWhisper says it can "Process sensitive content locally without data ever leaving your Mac" ⁴. That is the vendor's claim. We have not tested it, so treat it as a claim until you have run the check below.

WhisperType is a different case. It downloads its Whisper models from Hugging Face when you choose Download, and it can instead use a model file you import, which works offline ⁵. The download is a one-time network request for the model, and the import route avoids it entirely.

Four checks tell you which path an app takes

1. Look for an API key field. If the app asks for an OpenAI or other provider key, your audio goes to that provider under your account.
2. Read the settings for local and cloud modes. Find out which mode is the default, and whether a cloud model or AI cleanup is switched on.
3. Run the airplane-mode test on a real recording of a few minutes. Text that still appears with the network off was produced on the device.
4. Read the privacy policy, not only the store label. A 2023 study found that 97 percent of apps labelled Data Not Collected had privacy policies indicating otherwise ⁶. The label is a starting point, not proof.

A local default can change after an update

The same app can be local on one plan and cloud on another, and a setting can switch the path on later. An app that was local last month can add a cloud feature in an update. So a local default is not a permanent promise. Recheck the settings after each update, and rerun the airplane-mode test if the app's behaviour changes. Our question on whether AI dictation apps send your words to a language model lists the cleanup features that most often send text out.

Local Whisper and the OpenAI API differ in cost, setup and privacy

For the retention rules on the hosted route, read whether OpenAI stores Whisper audio.

What to do: choose the path, then check the one you got

1. For sensitive audio, choose a local app or the open-source model, and confirm it with the airplane-mode test.
2. If you choose an app that uses your own key, read the provider's data terms and your account's settings, not only the app's page.
3. Write down the mode you chose and the date, and update the note when the app changes.
4. Recheck after each update, since a feature can switch the path on.
5. For regulated records, read whether Whisper AI is HIPAA compliant before you dictate anything covered by a privacy rule.

For the wider picture of Whisper's privacy options, read whether Whisper AI is safe, and the overview of speech to text privacy for every platform. If you want a dictation app for Mac and Windows that works on the device, Yap, which we make, is one option. Read its privacy terms with the same checks you would apply to any app on this list.