Apple Watch Audio Intelligence Enters Developer Beta Testing

TechCrunch reported on November 2, 2026 that Apple’s Audio Intelligence features for Apple Watch are rolling out to developers in the iOS 27.3 beta, including Live Rewind and Siri Recap. A broader public beta is planned before the features become generally available.

Apple introduced the package in its September 9, 2026 Apple Watch Series 12 announcement, committing then to bring the two conversation features to beta later in the year. Apple’s support guide defines Audio Intelligence as four opt-in capabilities. Alongside Live Rewind and Siri Recap are Sound Recognition, which can notify wearers about sirens, alarms and doorbells, and automatic Music Recognition with Shazam, which displays a detected song’s title and artist in the Smart Stack.

Two Ways to Revisit a Conversation

Live Rewind is activated only by a deliberate double-press of the Digital Crown. It displays the previous 15 seconds of speech as text on the watch, with options to ask Siri about the content or save the snippet to the Siri app. Unless the user takes one of those actions, the text disappears approximately 30 seconds after the watch display dims.

Activation also plays an audible chime, even when the watch is silent or headphones are connected, and shows a full-screen animation and microphone indicator. Those signals are intended to alert people nearby that the feature has been used.

Siri Recap can run manually or on schedules based on time and location, and Control Center can turn it off at any time. It produces a title, brief summary and key points rather than a verbatim transcript, without attributing statements to identified speakers. Unsaved recaps auto-delete after seven days. Saved material can be edited, deleted or exported to apps including Notes and Journal. Apple cautions that generated summaries and snippets may be incomplete or inaccurate, and that important details may be omitted, misunderstood or summarized incorrectly.

A Hardware-to-Cloud Processing Chain

Apple’s Audio Intelligence Privacy Overview describes the S11 chip’s Secure Exclave as a hardware-isolated compartment that processes sensor data separately from the rest of the system. Its rolling audio buffer is inaccessible to watchOS, apps, the user and Apple, and old audio is continuously overwritten rather than accumulated as a recording.

For Siri Recap, a lightweight model on the S11 chip detects whether a conversation is occurring. The watch encrypts the audio and transfers it to the paired iPhone’s Secure Exclave, where on-device speech recognition transcribes it and a language model condenses the transcript to less than half its original length. The raw audio is then permanently deleted.

The condensed text is encrypted and sent to Private Cloud Compute, where Apple Foundation Models generate the title, summary and key points. Apple says the results are immediately deleted from the cloud service after being returned. Saved recaps and Live Rewind snippets can sync through iCloud with end-to-end encryption when two-factor authentication and a device passcode are enabled; Apple does not hold the encryption keys.

Siri Recap is also designed to omit financial data, government-assigned identifiers, authentication data and certain personal identifiers. The support guide cautions that automated filtering may not remove all sensitive information.

Compatibility and Initial Limits

Live Rewind and Siri Recap require Apple Watch Series 12 or Apple Watch Ultra 4 paired with an Apple Intelligence-enabled iPhone 16 or later, excluding iPhone 16e. Apple’s launch terms make English the starting language, state that the features will not initially be available in the EU, and subject features relying on server-side models to daily usage limits. The conversation features are unavailable to users under 13, and users under 18 with a Child Account need a parent or guardian to enable access.

The result is a specific kind of wearable AI: conversation recall built across watch, phone and cloud processing, with strict deletion and access controls but no promise of a faithful record. For users, the distinction that matters is between keeping raw audio inaccessible and trusting the generated text; the architecture addresses the first, while Apple’s own accuracy warning qualifies the second.