The update introduces Siri Recap and Live Rewind to the Apple Watch Series 12 and Ultra 4, following their initial announcement in September.
Siri Recap summaries are automatically deleted after seven days unless users choose to save them.
According to Apple, the features do not attribute speech to individual speakers, and Siri Recap is designed to exclude certain sensitive information.
Particular concerns relate to whether temporary audio processing or AI-generated summaries constitute interception under applicable legislation, and whether people nearby have meaningful opportunities to consent.
The beta release moves Apple’s conversation processing features from announcement to testing, allowing their technical safeguards and practical operation to be assessed.
4 Nov 2026
Apple begins testing AI-powered conversation processing on Apple Watch
Apple’s Audio Intelligence features, which allow smartwatches to process nearby conversations and generate text snippets or summaries, are now available for testing through the watchOS 27.3 developer beta, released on 3 November 2026. The update introduces Siri Recap and Live Rewind to the Apple Watch Series 12 and Ultra 4, following their initial announcement in September.
According to Apple’s technical documentation, Live Rewind allows users to retrieve the previous 15 seconds of speech as text by double-pressing the Digital Crown, while Siri Recap automatically generates summaries of conversations when enabled. Both features are optional, and users can control when and where they operate. Siri Recap summaries are automatically deleted after seven days unless users choose to save them.
Apple states that the features do not create or retain raw audio recordings. In its Audio Intelligence Privacy Overview, the company explains that Live Rewind processes audio within a protected buffer that continuously replaces older audio, while Siri Recap uses an AI model to detect nearby conversations before processing speech within the watch’s Secure Exclave. A condensed version of the conversation transcript is subsequently sent to Apple’s Private Cloud Compute infrastructure to generate a summary.
According to Apple, the features do not attribute speech to individual speakers, and Siri Recap is designed to exclude certain sensitive information. Live Rewind also provides audible and visual notifications when activated, while saved text snippets and summaries can be synchronised through iCloud using end-to-end encryption.
In initial testing reported by The Verge, Live Rewind generally worked as described in a quiet environment, although it occasionally misinterpreted words. The reviewer also found that Siri Recap identified the main topics of a work meeting and omitted some sensitive information, but that the resulting summary was less useful for her purposes. The review noted the difficulty of informing people nearby when Siri Recap is active and emphasised that further testing would be needed to assess the features’ performance and implications for bystander privacy.
The beta release follows earlier scrutiny of the features’ implications for privacy and consent. As previously reported by Digital Watch, legal experts have questioned how existing US state eavesdropping and consent laws apply to technologies that process conversations without retaining conventional audio recordings. Particular concerns relate to whether temporary audio processing or AI-generated summaries constitute interception under applicable legislation, and whether people nearby have meaningful opportunities to consent.
Why does it matter?
The beta release moves Apple’s conversation processing features from announcement to testing, allowing their technical safeguards and practical operation to be assessed. Although Apple has introduced measures to limit data retention and restrict access to audio, these protections do not necessarily address the privacy implications of generating and storing information derived from conversations. In particular, Siri Recap can produce summaries without notifying other participants, raising questions about transparency and whether individuals have meaningful opportunities to consent to the processing of their speech.
As previously reported by Digital Watch, the absence of a stored audio recording does not necessarily determine whether temporary speech processing, transcription, or summarisation falls within existing eavesdropping and consent laws. The distinction between the device owner’s decision to activate a feature and the rights of other conversation participants remains particularly important.
More broadly, the development illustrates how AI-enabled wearables are expanding the possibilities for ambient information processing, potentially challenging legal frameworks designed around more conventional forms of audio recording and surveillance. How these technologies are deployed, and how regulators and courts interpret existing protections, could influence future requirements for transparency, consent, and the processing of information.
