Apple Watch Audio Intelligence Beta: New AI Conversation Features for Series 12 and Ultra 4

Apple has officially launched the beta testing phase for new “Audio Intelligence” features on the Apple Watch Series 12 and Ultra 4, integrated as part of the iOS 27.3 deployment. These tools, which were first announced at a hardware event in September, introduce four optional capabilities designed to assist with audio recognition and conversation management.

Watch Captures Missed Conversation Details

The new suite includes two features focused on verbal interactions: “Live Rewind” and “Siri Recaps.” Live Rewind allows users to view the last 15 seconds of a conversation as text on the watch face by double-pressing the Digital Crown. This is intended to help users capture instructions or details they may have missed. To ensure transparency, the watch emits a signal tone and displays an animation when this feature is active, alerting those nearby that audio is being processed. If not saved, these text snippets are automatically deleted after 30 seconds. Users may also see an option to “Ask Siri” beneath the text to inquire about discussed terms, and saved snippets are stored in the standalone Siri app introduced with iOS 27.

Siri Recaps provide summarized bullet points of conversations, extracting key information such as dates, deadlines, or pricing details. According to Apple, these summaries do not provide verbatim transcripts, name speakers, or store raw recordings. Users can review these notes at the end of the day to save them to the Notes app or the Journal app; otherwise, the data is automatically deleted after seven days. Apple suggests these tools are useful for meetings, parent-teacher conferences, or personal training sessions, though the system will not record conversations while the user is listening to music via headphones with noise cancellation enabled.

Data Processing and Security

The system relies on a hybrid processing architecture involving the new S11 chip, the paired iPhone, and Apple’s Private Cloud Compute (PCC). A protected audio buffer handles initial processing locally, while sensitive data is encrypted before being transmitted to the PCC. Apple states that the system is designed to exclude sensitive information, including government identification numbers, financial credentials, and content promoting self-harm.