Apple's always-listening Watch features keep raw audio inside a hardware Secure Exclave

Apple has made the Apple Watch an always-available listening device. At Wednesday's iPhone launch event, where a foldable iPhone also made its debut, the company introduced new Siri audio features for its watches that process live conversations on the wrist, TechCrunch reports. Apple simultaneously published a technical document explaining how the features protect privacy, and The Verge went through its details.

What the features do

Live Rewind is triggered by a double-press of the Watch's Digital Crown and reaches back fifteen seconds, producing a text transcript of what was just said. The result can be saved to Apple's new standalone Siri app and revisited later. Apple pitches it for capturing a book recommendation or a colleague's idea, but as TechCrunch notes, the same mechanism can catch someone saying something they never meant to have recorded.

Siri Recap goes further, using ambient listening to produce high-level notes of conversations. These are not exact transcripts: Apple says the feature generates a title, a summary, and key points that can be reviewed in the Siri app on iOS.

Sound Recognition, the least controversial piece, is aimed at deaf and hard-of-hearing users. On-device AI can alert the wearer to sirens, alarms, doorbells, or a crying baby, even when the paired iPhone is not nearby. The Verge also lists Music Recognition among the newly announced capabilities.

The privacy architecture

According to The Verge, Apple's document states that raw audio is handled inside dedicated hardware, never saved as a file, and unreachable by the operating system, apps, the user, or Apple itself. The S11 chip in the Apple Watch Series 12 and Apple Watch Ultra 4 includes a Secure Exclave, an isolated compartment built into the silicon that processes sensor data separately from the rest of the system. Microphone audio enters this compartment, where it is analyzed for speech, sounds, or music without transcribing or storing the raw signal; the buffer is continuously overwritten and never becomes a recording.

If processing needs to move to the iPhone, the data is transferred with encryption through the Secure Exclaves on both devices. Text that users choose to keep from Live Rewind and Siri Recap syncs with end-to-end encryption, as long as a device passcode and iCloud two-factor authentication are in place.

Controls matter too. Live Rewind requires an explicit double-press of the Digital Crown each time it is used, The Verge notes. TechCrunch adds that Siri Recap does not default to always-on operation the way some other AI wearables do; it can be scoped to specific contexts such as work hours or daytime, and toggled on or off from Control Center. Apple also says the features do not identify or attribute speakers, and that activation comes with an audible chime plus a full-screen microphone animation.

Consent, law, and culture

TechCrunch raises the unresolved questions. Other recording-focused companies already treat consent as the user's problem: Plaud advises customers to obtain legally required consent before recording, and Amazon's Bee terms place compliance with laws around minors' data on the user, not Amazon. Apple's text-only output adds a new wrinkle, because it is unclear how courts would treat transcripts with no underlying audio to authenticate them, and the evidentiary rules are likely to vary by state.

There is also a cultural cost. Studies have shown that awareness of recording changes how people behave and express themselves, and grabbing fifteen seconds of audio via a watch crown is a subtler gesture than holding up a phone. TechCrunch also speculates that Apple may have moved now not because consumers were demanding it, but out of concern that a rival such as OpenAI would ship a competitive AI device first.

Why it matters

Apple is betting that hardware-level guarantees, namely a Secure Exclave, no stored audio, and end-to-end-encrypted text, can make ambient listening acceptable on a mainstream device worn by millions. If the bet pays off, always-on audio AI graduates from niche pendant startups to the default wrist. If it backfires, Apple has attached a surveillance-adjacent capability to a product line marketed heavily on privacy. Either way, this rollout sets the template for how consumer devices handle raw audio from here on.