Apple’s own one-sentence definition of Audio Intelligence names cloud models alongside on-device ones. Support document 148354 calls it a brand-new class of Apple Intelligence features that use the microphones of Apple Watch, the power of Apple silicon, and on-device and cloud-based AI models to help you detect sounds, identify music, and stay present and engaged, all while protecting your privacy at every step.
That phrasing sits next to a separate claim about the S11 chip’s Secure Exclave, which Apple describes as handling audio processing on the device. Apple does not map the two statements onto each other, so the honest account of what happens to what is shorter than most readers will expect.
Key facts
| Item | What Apple published |
|---|---|
| Source | Apple support document 148354, September 9, 2026 |
| Models named | On-device and cloud-based AI models |
| Hardware named | Apple Watch microphones, the S11 chip, the Secure Exclave |
| Secure Exclave | A dedicated hardware-isolated compartment used for on-device audio processing |
| Storage claim | Apple says no audio recordings are stored |
| Usage limits | Daily usage limits apply to Apple Intelligence features that rely on server-side models |
| Per-feature split between on-device and cloud | Not published |
| Product-specific privacy notice for Audio Intelligence | Not published |
What Apple states about the processing
Two Apple statements are in play, and both are short. The support document names on-device and cloud-based AI models as inputs to the category as a whole. The Apple Watch Series 12 page says Audio Intelligence uses the built-in microphone and secure S11 chip in Series 12, alongside the power of Apple Intelligence, to help you make sense of what you hear in a private and secure way.
Don’t miss the best of The Mac Observer
Set us as a preferred source and our Apple reporting ranks higher in your Google Search results and Discover feed — one tap, no account changes.
On privacy, Apple names one mechanism: the Secure Exclave, which it describes as a dedicated hardware-isolated compartment. Apple says no audio recordings are stored.
Everything past that is absent rather than reassuring or alarming. Apple does not say which of the four features uses a cloud model, what leaves the watch, whether anything leaves the watch at all, or what form it takes if it does.
Documented against undocumented
| Apple documents this | Apple does not document this |
|---|---|
| On-device and cloud-based AI models are both used by the category | Which of the four features uses which |
| A Secure Exclave on the S11 is used for on-device audio processing | What the Exclave holds, or for how long |
| No audio recordings are stored | Whether any audio or text is transmitted, and to where |
| Daily usage limits apply to server-reliant features | Any number for those limits, for any feature |
| Live Rewind produces a text snippet of the previous 15 seconds | How that snippet is produced, held or discarded |
| Siri Recap produces high-level summaries of conversations | Where a summary is generated or stored |
The daily limits are the one hard clue
Apple’s Apple Intelligence footnotes say certain features that rely on server-side models are subject to daily usage limits, and that those limits may vary by feature, request complexity and system demand. The same footnotes add that expanded access to such features will be available for a fee in the future.
Audio Intelligence is presented as an Apple Intelligence category and its support document carries the usage-limit language. That is Apple placing at least part of the set in the same bracket as its other server-assisted features, which is a documented statement rather than a deduction about audio.
What Apple’s general privacy pages add
Apple’s privacy features page describes Apple Intelligence generally, not Audio Intelligence specifically. It says Apple Intelligence analyzes whether a request can be processed on the device or if it requires Private Cloud Compute, that data is never made accessible to Apple, and that independent experts can inspect the code that runs on the servers.
For Siri AI it adds that server-assisted searches and requests are associated with a random identifier rather than an Apple Account. None of that language names Apple Watch, the S11 or any of the four Audio Intelligence features, and Apple has published no product-specific privacy notice for the Series 12 and Ultra 4 feature set.
What Apple has not said
- Which Audio Intelligence features use cloud-based models and which do not.
- Whether Private Cloud Compute is the cloud path involved, since the support document does not name it.
- What the Secure Exclave holds while a feature runs, or what deletion means in practice.
- Any figure for the daily usage limits, for Audio Intelligence or for anything else.
- Whether a paired iPhone performs any part of the processing.
- Any audit, report or technical paper covering the Exclave.
watchOS 27 arrives free on Monday, September 14, the watches reach stores on Friday, September 18, and Live Rewind and Siri Recap are dated only as a beta later in 2026. If Apple publishes a support document that maps the four features to on-device or cloud processing, that is the page to read; our event recap covers the rest of September 9.
