Apple Watch’s AI features for rewinding and recapping conversations begin beta testing

Apple has moved its AI-powered Audio Intelligence features for Apple Watch into beta testing, starting with developers through the iOS 27.3 update. The company first announced the technology at its Surprise and Shine hardware event in September. The suite consists of four opt-in features, aimed at Apple Watch Series 12 wearers, that detect important sounds, identify songs, catch up on what was just said, and take notes on conversations.
Two of the features are accessibility and utility oriented. Sound recognition helps deaf and hard-of-hearing users identify sounds such as sirens, alarms or doorbells from the watch. Song recognition uses Apple's Shazam app to detect music playing nearby and return the track name and artist.
The more sensitive pair, Live Rewind and Siri Recaps, listen to human speech. Live Rewind is built for the moment you miss something: a set of verbal directions, or a long list of restaurant specials. Double-pressing the watch's digital crown displays the last 15 seconds of speech as a text snippet on the watch face. If you do nothing, the snippet disappears permanently. If you want to keep it, you can save it to the standalone Siri app that shipped with iOS 27, which could help with remembering a name or capturing a note from a fast-moving presentation. A button beneath the snippet labeled "Ask Siri" lets users pose follow-up questions about what was said, such as asking about an unfamiliar technical term or a sauce a waiter mentioned.
Apple has built a visible consent cue into the gesture. A chime sounds and an animation plays when the crown is double-pressed, signaling through gesture, sound and display that something was captured. Snippets are deleted automatically after 30 seconds if not saved.
Siri Recaps goes further, producing high-level notes of a conversation. Per Apple, the feature does not identify speakers, retain raw audio or produce word-for-word transcripts. Instead it condenses a conversation into a short summary sentence plus bullet points covering key points, which may include dates or deadlines, pricing or payment plan details, and other factual information. Apple describes the intended use as mimicking a person's own note-taking without adding emotion or tone, and points to settings such as work meetings, parent-teacher conferences and personal training sessions. Recapping does not run while the wearer is listening to music through headphones or using noise cancellation.
Recaps are reviewed rather than silently stored. When the feature is enabled, users are prompted at the end of the day to go through saved recaps and choose which to keep; they can be saved to the Notes app or exported to Journal. Anything not saved is deleted within seven days. The review prompt also appears when the watch is first unlocked in the morning. Users can turn recapping on manually or schedule it, for example only during work hours or never at night, and can switch it off at any time from Control Center. Apple says the feature is designed to omit sensitive material, including government-assigned identifiers, financial and authentication data, certain personal identifiers, and language promoting self-harm.
On the privacy architecture, the features rely on a combination of hardware, software and services. Lightweight models on the watch's S11 chip detect whether a conversation the wearer is part of is taking place. The always-on processor includes a secure enclave, a hardware-protected memory region inaccessible to the rest of the operating system, where a dedicated audio buffer handles raw audio. That audio is reachable only by the paired iPhone through a secure pairing mechanism. On-device processing on the iPhone condenses the audio into text with filler words stripped out; that text is encrypted and sent to Apple's Private Cloud Compute to generate the recap, which returns to the phone; the recap and the summarization inputs are deleted from PCC immediately afterward. Recaps are accessible only to the user, and encrypted when synced across devices. Live Rewind works the same way, with the watch handing audio to the iPhone for processing. Apple says the features should not affect battery life on the watch or phone. A public beta is planned before general availability.
This suggests Apple is treating ambient listening as a hardware-enforced trust problem: the enclave, the phone-side processing, the short-lived cloud step, and the delete-by-default windows are all designed so that even Apple cannot reach the audio. In practice, the chime-and-animation cue and the no-headphones rule are the parts most likely to matter in shared spaces, because they make capture observable to bystanders. A likely trade-off is that summarizing rather than transcribing limits how much detail users recover, especially compared with dedicated AI recorders that keep full transcripts. It also means the initial release is confined to Series 12 and Ultra 4 hardware, which leaves older watches out.
Why it matters: developers and product teams planning watch or wearable experiences now have a documented model for on-device detection, secure handoff and cloud summarization with aggressive retention limits, which is a useful pattern to copy. If the beta holds, expect user-facing expectations around visible recording cues to harden across the AI wearable category, as inference from Apple's design choices. Because the rollout starts with developers in iOS 27.3, teams can begin testing integrations well before public availability.
Based on reporting from the original publisher. Visit the source for full context and later updates.
Publisher excerpt
Now in beta: Apple's new Audio Intelligence features can help you remember what someone said or take notes on conversations.