Live Rewind keeps a short rolling buffer of about 15 seconds, converts it to text on the watch, and overwrites the buffer without permanently storing recordings.

Siri Recap converts speech to text to generate a brief summary with a title and key points, then deletes the original audio and processes data using Private Cloud Compute to protect identity and speakers.

Audible and visual indicators accompany Live Rewind to signal when listening is occurring, including a full-screen animation and a microphone icon for privacy transparency.

Apple unveils Audio Intelligence for the Apple Watch Series 12 and Ultra 4, introducing Siri Recap and Live Rewind, with on-device processing, privacy safeguards, opt-in controls, and Shazam integration.

The company says Live Rewind and Siri Recap operate on-device, do not create conventional audio recordings, and will be released in beta later this year alongside health sensing upgrades and a faster S11 chip.

Both features can be disabled, and users can control when and where they operate; processing occurs in secure hardware isolated from the OS, Apple, and third parties.

The watches, powered by an upgraded S11 processor, can continuously buffer audio around the wearer and generate transcriptions for later reference, with transcriptions and data treated as private.

There is no saved audio file on device or cloud; buffering replaces old audio every 15 seconds, and data generally disappears after around 30 seconds for Live Rewind or seven days for Siri Recap unless saved by the user.

User access to rewind is via double-tapping the crown and holding the watch to the ear or using Bluetooth headphones, delivering the last 15 seconds of audio and potential transcription.

Privacy and legal scrutiny in the United States address whether continuous on-device speech analysis could run afoul of state eavesdropping or wiretapping laws that require consent from all parties in a private conversation.

Experts suggest these privacy controls could influence future smart glasses strategies, potentially requiring visible or audible notices when listening or recording to engage nearby observers.

The feature set aims to help with remembering names and quick facts from conversations, as well as other common memory-challenge moments in daily life.