Apple has started beta testing Audio Intelligence on Apple Watch, TechCrunch reported on 2 November, initially for developers in the iOS 27.3 update. Two of its four opt-in features listen to conversations. Live Rewind shows the last 15 seconds of speech as text. Siri Recap takes notes on conversations through the day, and the recaps auto-delete after 7 days unless the user saves them, Apple’s support page says.
Apple’s privacy overview lists six steps a Siri Recap passes through. By our count, Apple says no one can access the data at four of them and only the user at the other two; Apple itself is not listed at any step. Live Rewind plays a chime for people nearby. Siri Recap, Apple says, has no audible signal, because no raw audio is retained.
The features work only on Apple Watch Series 12 and Apple Watch Ultra 4, paired with one of 12 iPhone models Apple lists, our count of its support page. Apple announced them in September, TechCrunch says. On its 2 November earnings call, chief executive John Ternus said Audio Intelligence is “coming in a future software update”.
The key insight: By our count of Apple’s own table, Apple says no one can access a Siri Recap conversation at four of its six steps and only the user at the other two. Live Rewind plays a chime for people nearby; Apple says Siri Recap has no audible signal because no raw audio is retained.
Audio Intelligence has four features, Apple says. Sound Recognition detects sounds such as sirens, alarms and doorbells. Music Recognition with Shazam shows the song playing in the background in the Smart Stack. Live Rewind and Siri Recap are the two that handle speech.
Live Rewind is triggered by a double press of the Digital Crown. Apple says it “only shows the last 15 seconds of speech as a text snippet on your display” and “doesn’t continue to transcribe in the background”. The user can ask Siri about the snippet or save it to the Siri app.
Siri Recap “ambiently takes notes on your conversations throughout your day”, Apple says, and creates “high-level Apple Intelligence–generated summaries”. Users must opt in, and Apple says “Siri Recap runs only when you choose”: on a schedule by time or place, such as “only at work” or “never at night”, or switched on and off by hand from Control Center.
Apple says Siri Recap “doesn’t produce a transcript” and does not attribute speakers. Summaries are designed to omit sensitive information such as “financial data, government-assigned identifiers, authentication data, and certain personal identifiers”. Apple’s footnote warns that “automated filtering may not remove all sensitive information”.
Live Rewind and Siri Recap need Apple Intelligence and Siri AI (Beta), start in English, and are not available for users under 13. Apple’s support page says Siri AI “will not initially be available in the EU on iOS, iPadOS, and watchOS”.
Who can access a conversation at each of the six steps of Siri Recap, as labelled in Apple’s Audio Intelligence Privacy Overview (September 2026). Our count: No one at four steps, Only you at two. The labels are Apple’s claims; we have not tested them.
How Siri Recap Moves a Conversation
A lightweight model on the watch’s S11 chip checks whether speech is happening, Apple says. If it is, audio flows into a protected buffer in the chip’s Secure Exclave, then travels encrypted to the Secure Exclave on the paired iPhone, where it is transcribed on the device.
An on-device model then writes a condensed version “less than half the length of the original transcript”, and the raw audio is deleted. That condensed text goes, encrypted, to Apple’s Private Cloud Compute, where Apple Foundation Models write the title and summary.
Context travels with it. Apple lists Now Playing data, calendar data, high-level location labels such as home, work and school, and city, state and country. “Precise location and specific points-of-interest are not included”, the overview says. Processing in Private Cloud Compute is “subject to daily usage limits”.
Apple’s table of the journey labels each step by who can access the data. The buffered audio on the watch and the phone, the speech-detection stream and the text inside Private Cloud Compute are all marked “No one”. The condensed text on the iPhone and the finished summary are marked “Only you”.
The Signal for People Nearby
For Live Rewind, Apple says, “an audible chime plays from the speaker on your Apple Watch, even if your Apple Watch is on silent or you have headphones connected”. A full-display animation and a microphone indicator appear. “This is to alert those nearby that the feature has been activated”.
Siri Recap works differently. The overview says: “There is no audible signal because no raw audio is retained”. Apple adds that “there is no way to reconstruct the original audio from a Siri Recap or share raw audio with anyone”.
TechCrunch put the design in context: “Given the backlash to AI recording devices, like Meta’s AI glasses and other AI wearables, Apple has attempted to carve out a space where the features can be useful to Apple Watch owners, while still respecting the privacy of those around them”.
Upgrades and Cloud Costs
The features need new hardware on the wrist and a recent iPhone. Apple says only the S11 chip’s Secure Exclave makes them possible; its overview states that “Audio Intelligence is only possible because of the Secure Exclave on our new S11 chip”. Finance chief Kevan Parekh said on the call that “AI offers a compelling reason for customers to upgrade and switch”.
TechCrunch reported that Apple says none of the features should affect battery life on the watch or phone, and that they will go to a public beta before a general release.
Business Pill · Consent
An evergreen Business Engineer explainer on consent: you can only agree to what you know. It teaches the general idea only and does not discuss Apple, Apple Watch or Audio Intelligence.
As we read Apple’s documents, the bystander signal goes to the feature the user triggers for 15 seconds at a time. The feature that runs through the day relies on its design instead: no recording, no transcript and no speaker attribution, by Apple’s account.
As we read it, the privacy design is also a hardware requirement. Apple ties all four features to the Secure Exclave of the S11 chip, so they run only on Apple Watch Series 12 and Apple Watch Ultra 4.
Apple, Audio Intelligence Privacy Overview, September 2026
“There is no recording to share, forward, or produce if requested by any party, because no recording exists”
Three Implications
FOR PEOPLE NEARBY Live Rewind plays a chime and shows a microphone indicator when used. Apple says Siri Recap has no audible signal because no raw audio is retained.
FOR UPGRADES Audio Intelligence needs an Apple Watch Series 12 or Ultra 4, and its two speech features need one of 12 iPhone models Apple lists, our count. Parekh said AI offers a compelling reason for customers to upgrade and switch.
FOR APPLE’S CLOUD Siri Recap summaries are written by Apple Foundation Models on Private Cloud Compute, which Apple’s overview says is subject to daily usage limits.
The framework observes that model capability and product capability “are not the same curve, and the gap between them — what the platform makes possible versus what the product surfaces — has a name.” It calls that gap overhang.
As we read it, Siri Recap shows a gap set on purpose. Apple says the iPhone transcribes the speech on the device, yet the product surfaces only a short summary, with no transcript and no speaker labels. Here the gap comes from Apple’s privacy design rather than from model capability. The framework was published in June 2026 and does not discuss Apple Watch.
What Is Not Established
The start of beta testing and the iOS 27.3 version rest on TechCrunch’s report; we did not read Apple release notes for that beta. Apple’s support page, dated 30 September, says the two features will arrive in beta “later this year”. Apple has not given a date for the public release in the documents we read.
Every access label above is Apple’s own claim; we have not tested it. Apple says “independent security experts can verify this promise at any time” for Private Cloud Compute. TechCrunch wrote that Live Rewind snippets delete after 30 seconds; Apple’s documents say the text disappears 30 seconds after the watch display dims.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growthstory. This week it started being priced as a credit story.
Apple is testing an Apple Watch feature that takes notes on conversations all day, with recaps deleted after 7 days unless saved. Apple’s own map of the data lists no one, not Apple, at four of six steps and the user at two. Live Rewind signals people nearby with a chime; Siri Recap, by Apple’s design, does not.
A note on sourcing. The start of beta testing is from TechCrunch’s report of 2 November 2026. Feature details, requirements and privacy claims are from Apple’s support page for Audio Intelligence and its Audio Intelligence Privacy Overview of September 2026, both read in full. The two call quotes are from Apple’s 2 November webcast replay, transcribed by machine. The step count is our count of Apple’s table; the access labels are Apple’s claims, not tested by us. We did not contact Apple.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
