Siri Recap on Apple Watch Ultra 4 Produced 50+ Summaries, but Only a Few Were Useful
Siri Recap on the Apple Watch Ultra 4 listens to nearby conversations and turns them into a short summary, so you don't have to take notes during a meeting. The Verge reviewer generated more than 50 summaries but called only a few useful, mostly work-related ones. On the other hand, 90% of the texts are unintelligible to anyone but him, so the leak risk is nearly nil. The takeaway: strict privacy reduces risk, but AI quality drops as well.
AI-processed from The Verge; edited by Hamidun News
Imagine a watch that listens to the conversations around you all day and then sends you a short digest: what was said at a meeting, what was discussed at a parent-teacher conference. That is exactly what Siri Recap, a beta feature of the Apple Watch Ultra 4, does. The Verge reviewer wore it for two weeks while also testing the Amazon Bee Frontier, a similar wearable AI gadget.
He ended up with more than 50 summaries (each one a short digest of a single conversation) and found only a few useful, mostly work-related ones. By his estimate, 90% of the summaries would be meaningless to anyone but himself. It is almost a lab example of how data protection strips AI of its usefulness.
How Apple Protects Conversations
Bee records everything and stores transcripts. Apple chose a minimalist approach focused on privacy. Siri Recap is off by default. The user turns it on and sets a schedule, for example only at the office from 13:00 to 14:00 on weekdays. Summaries are deleted after one, three, or seven days unless the user saves them manually.
Now for the engineering part. Raw audio is never stored. Sound is processed inside a secure enclave on the 2026 watch's S11 chip and deleted immediately. Neither Apple nor third parties have access to the enclave. Most processing happens on the device, and anything that does go to the cloud passes through Private Cloud Compute. Because of hardware requirements, the feature will not run on older Apple Watches.
The summaries contain no transcript showing who said what. That makes them hard to use as "evidence" in an argument, in court, or for blackmail. The flip side: the feature has no external recording indicator, like Live Rewind, meaning no signal and no sound. You will have to warn your conversation partners yourself and, most likely, answer questions about how the whole thing works.
Where Privacy Hurts Quality
Siri Recap does not learn the user's voice. That is a plus for privacy, but it makes it hard for the model to tell when the person is actually talking and when a sound is simply playing nearby. The reviewer played a TikTok video about a blogger who was taken to the hospital with sepsis. The summary made it sound as though he himself, or someone he knew, had ended up in the hospital. Another clip, a breakdown of a Reddit story, turned into a summary about some Dana and sparkles on another woman's husband. Without indicating who is speaking, such recordings come out funny and useless.
There is a workaround: Siri Recap switches itself off when you wear AirPods. But wearing AirPods just so the feature doesn't get confused means adjusting your own behavior to fit its limitations.
The second problem: the AI doesn't know what a person needs to remember. Siri Recap focuses on facts: numbers, dates, lists. At a work meeting, that helps. During a briefing on new smart glasses from Google, Samsung, Gentle Monster, and Warby Parker, the summary reminded him of the main topics, and some of the interviews were tied to files in Voice Memos. But in one interview, the most interesting parts were the speaker's personal story and his sharp opinions, and neither made it into the summary. Meanwhile, a 20-minute argument about wired versus battery-powered smoke detectors was broken down point by point, even though he would have preferred to forget it.
The Bottom Line: Usefulness Versus Cost
Out of more than 50 summaries, the author judged only a few to be useful. The upside is that 90% of the recordings are understandable only to him, so the leak risk is low. Amazon Bee proved more useful because it stores full transcripts. But after seeing them, the author decided that living that way was not worth it, and deleted all of his data.
The main cost turned out to be social. Before every conversation, you have to decide whether to warn people, explain how the feature works, or ask colleagues for consent. There are no common norms for this yet. Turn the feature off, and you don't have to think about any of it at all.
What this means for developers and product managers: first, on-device processing and not storing audio do reduce risk, but summary quality falls as a result. Without speaker recognition and context, the model confuses the speaker, the background, and the participants in the conversation. Second, a summary "by facts" is not the same as a summary "by meaning": users want their personal emphases, while the model picks out the numbers.
Third, a consent interface, meaning a recording indicator and clear settings, becomes part of the product rather than a legal footnote. Siri Recap is still in beta, and these conclusions rest on one reviewer's experience, but the tension between privacy and usefulness will persist for the next devices of this type.
- Nov 21, 2026OpenAI GPT-6 Cyber Found About 35 Flaws in iPhone, Mac and Other Apple Devices
- Nov 21, 2026Apple's macOS 27.3 beta 2 will let you limit AI agents' access to Mac files
- Nov 21, 2026Apple Is Building an AI Agent for Mac While Making It Harder for Outsiders to Get In: 14 Clicks Instead of 4
- Nov 21, 2026Apple Is Recruiting 200,000 Volunteers to Test a Health AI Chatbot
Key Questions Answered
How Apple Protects Conversations?
Bee records everything and stores transcripts. Apple chose a minimalist approach focused on privacy. Siri Recap is off by default. The user turns it on and sets a schedule, for example only at the office from 13:00 to 14:00 on weekdays. Summaries are deleted after one, three, or seven days unless the user saves them manually. Now for the engineering part. Raw audio is never stored. Sound is processed inside a secure enclave on the 2026 watch's S11 chip and deleted immediately.
Where Privacy Hurts Quality?
Siri Recap does not learn the user's voice. That is a plus for privacy, but it makes it hard for the model to tell when the person is actually talking and when a sound is simply playing nearby. The reviewer played a TikTok video about a blogger who was taken to the hospital with sepsis. The summary made it sound as though he himself, or someone he knew, had ended up in the hospital. Another clip, a breakdown of a Reddit story, turned into a summary about some Dana and sparkles on another woman's husband.
What are the key details on The Bottom Line: Usefulness Versus Cost?
Out of more than 50 summaries, the author judged only a few to be useful. The upside is that 90% of the recordings are understandable only to him, so the leak risk is low. Amazon Bee proved more useful because it stores full transcripts. But after seeing them, the author decided that living that way was not worth it, and deleted all of his data. The main cost turned out to be social. Before every conversation, you have to decide whether to warn people, explain how the feature works, or ask colleagues for consent.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.