With the launch of Siri AI, Apple raises an additional question: Will voice recordings and text be allowed to be used in the training of its own language models in the future? The collected data remains the same – what changes is its intended use.
Apple published a support document on September 14th regarding both settings, coinciding with the release of the new operating system series. In it, the company separates two options for the first time: the existing setting for Siri and dictation, and a new one that additionally covers the training of the Foundation Models. Which of the two appears on a device depends on whether Siri AI is even running there – and that is precisely the crucial question in this country.
Key Facts at a Glance
- The new sharing option appears when setting up Siri AI, with the options to share or decline audio and text.
- The same data will be collected as before: excerpts from Siri queries, dictation and translation including audio, transcripts and associated query data.
- What's new is the use – the data can now also be used to train the Foundation Models.
- Those who previously agreed to the old permission will be asked again when activating Siri AI.
- Siri AI will not launch on iPhone, iPad, and Apple Watch in the EU. The existing option will remain in place there.
What is the difference between the two settings?
| Previous release | New release | |
|---|---|---|
| Who gets to see them? | all users | Siri AI users |
| Which data | Audio, transcripts and query data from Siri, dictation and translation | the same |
| What is it used for | Siri, Dictation and related voice functions | additional training of the Foundation Models |
| Can be changed later | Yes | Yes |
Apple describes the data assignment in the same way in both cases: it is linked to a randomly generated device identifier, not to an Apple account or an email address. Audio recordings are listened to exclusively by Apple personnel; a portion of the remaining samples is sent to reviewers. The data is not used for advertising profiles and is not sold to third parties.
Why this doesn't matter on German iPhones for now
The new permission is explicitly tied to Siri AI. Since the assistant won't initially launch in the EU on iOS, iPadOS, or watchOS, the prompt won't even appear on iPhones. The existing options for Siri and the dictation function remain unchanged.
The situation is different on Mac and Apple Vision Pro: Siri AI also runs in the EU there, and consequently, the new question appears during setup. Anyone trying out the assistant in this country is therefore simultaneously deciding whether to allow training – and doing so on a device where they might not expect it.
Where the attitude lies
You can change the sharing settings at any time via the privacy settings in the Analytics & Improvements section. Apple describes the process separately for iPhone, iPad, Mac, Apple Watch, HomePod, Apple TV, and Apple Vision Pro.
Two special cases are worth mentioning: On the Apple Watch, the setting inherits the configuration from the iPhone it was set up with. If someone sets up a watch for a family member, sharing is disabled by default – and cannot be changed there. The same logic applies to the HomePod, and to the Apple TV only if the setup was done via an iPhone or iPad.
A consent that covers more than before
The distinction is more subtle than it first appears. Previously, the focus was on improving an assistant that runs on the user's own device. Now, it also includes models that Apple trains across device and user boundaries, models that end up in functions unrelated to Siri.
Apple's decision not to silently continue the old consent but to ask again is the right one – and the reason why the request is also encountered by people who believe they have already clarified this.
Practically speaking, the leverage remains limited as long as the assistant doesn't run on iPhones in this country. For Mac and Apple Vision Pro, however, it's worth checking the settings, because the permission is enabled during setup and can easily be overlooked.
Do you allow voice recordings for training Apple devices, or do you generally draw the line when it comes to audio? Let us know your decision in the comments.
