Apple says its new always-on audio features never record you. Here is how.
A dedicated security chip in the new Apple Watch processes your voice and sounds in a sealed compartment, then discards the raw audio instantly. Apple says even Apple cannot hear it.
Key points
- Apple announced four new Audio Intelligence features at its iPhone Duo event on 10 September 2026, including Siri Recap and Live Rewind.
- Raw audio from these features is processed inside a dedicated hardware compartment on the S11 chip and is never saved as a file.
- Apple says the company itself cannot access the audio, and neither can apps or the device's operating system.
- Users must actively trigger the Live Rewind feature every time; it does not run automatically in the background.
- Any text that does leave the watch is protected by end-to-end encryption, meaning only you can read it.
Apple Watch has a new party trick: it listens for speech, sounds, and music all day long. Naturally, the first question is who else can hear what it picks up.
At Wednesday's iPhone Duo launch event, Apple introduced a set of Siri features it calls Audio Intelligence. The four features are Siri Recap (a short summary of recent spoken content), Live Rewind (a replay of what was just said), Sound Recognition (identifying sounds like alarms or a baby crying), and Music Recognition (naming a song). Alongside the launch, Apple released a technical document explaining how it handles the audio those features collect.
So is the watch recording everything I say?
No, and the reason is a piece of hardware called the Secure Exclave. Think of it as a locked room built into the S11 chip, the processor inside the new Apple Watch Series 12 and Apple Watch Ultra 4. Audio enters that room, gets analysed, and is then overwritten by the next slice of audio, continuously, in a loop. No audio file is ever written to storage.
Apple's document states the raw audio "is not accessible to the operating system, apps, or Apple." The watch's own software, watchOS, cannot reach into that compartment. Neither can any third-party app. Neither can Apple.
The key phrase is "continuously overwritten stream." The watch is not silently building a recording. It is more like a whiteboard that is wiped every few seconds before anything is written down.
What actually leaves the watch?
Text does, but only if you choose to keep it. Siri Recap and Live Rewind produce a written summary or transcript, and users decide what to save. If that text syncs to your iPhone, it travels through an encrypted tunnel between the Secure Exclaves on both devices.
For that sync to work with iCloud, you need a device passcode and iCloud's two-factor authentication (a security setting that requires a second confirmation, like a code sent to your phone, before anyone can log in). If you have both switched on, the text is end-to-end encrypted, meaning Apple's servers carry it but cannot read it.
Live Rewind is not passive at all. You must double-tap the Digital Crown, the small dial on the side of the watch, every single time you want it to activate. It will not replay audio you did not ask it to replay.
What does this mean for me?
If you buy an Apple Watch Series 12 or Ultra 4 and switch on Audio Intelligence, you are not handing Apple a live feed of your conversations. The design Apple describes keeps audio inside dedicated hardware and throws it away before it can become a recording.
What Apple cannot guarantee is everything else: how well the chip actually enforces that boundary, or what future software updates might change. Independent security researchers have not yet audited the system. The document is a company claim, not a peer-reviewed finding. That distinction matters.
For now, the practical steps are simple: keep your device passcode active, keep two-factor authentication on for iCloud, and check which Audio Intelligence features you have enabled in settings.
