I tried Apple’s Siri Recaps. They’re surprisingly useful — and a little unsettling

Apple on Tuesday shipped the first developer betas of iOS 27.3 and watchOS 27.3, adding four Audio Intelligence features to testing on the Apple Watch Series 12. Two — important-sound detection and Shazam-powered song identification — are straightforward, with the former aimed at deaf and hard-of-hearing users. The other two, Live Rewind and Siri Recaps, are the controversial ones: they record, transcribe, and summarize conversations happening with and around the wearer.

Live Rewind retrieves what was said in the previous 15 seconds. Siri Recaps runs further, summarizing audio across the day and saving notes of important things said. Apple's stated privacy guardrails: audio and AI models processed on-device when possible, no speaker identification, no storage of raw audio files, and redaction of private data such as government ID numbers and personal account information.

The author's first day with the feature tested it in a concrete setting — dropping off a teen's car for repairs and collecting a rental, where employees rattled off time frames, prices, and policies too quickly to jot down. Siri Recaps later surfaced those details for reference. The author expects to lean on Recaps for saving notes, but doubts Live Rewind's appeal, predicting a spoken "sorry, what?" instead of reaching for the watch.

Notably, the author did not frame the judgment as settled: AI-assisted note-taking can also capture things not worth remembering, and areas for improvement remain. The broader unease — that these features normalize technology that is always listening — sits alongside the acknowledged value of what Apple built.

What to watch next: whether the privacy design holds up as these features move from beta to general release, whether Live Rewind finds any real use case, and whether the recap quality sharpens enough to justify a permanent listening surface on the wrist.

For builders, the episode is a template for shipping ambient-audio AI: on-device processing, no speaker IDs, no raw audio retention, and redaction of sensitive identifiers are the stated trade for earning a skeptical user's daily habit.

The unresolved question is where the line falls between a useful recap and surveillance — a question Apple has answered with defaults rather than proof, and one testers and regulators will keep pressing.