During its Sept. 9 keynote, Apple announced the Apple Watch Series 12 and the Ultra 4, both of which will come with a feature that passively listens to your audio conversations all day and can summarize them for you on demand. It's part of a collection of features called "Audio Intelligence," but the two key features to consider here are: Live Rewind: If you double-press the Digital Crown on your Apple Watch, it will pull up a text snippet of your conversation over the last fifteen seconds. Siri Recaps: Your watch listens to your conversations and generates an AI summary of any of your transcribed audio at the end of each day. At first glance, the basic concept seems like a privacy nightmare, but Apple has put some safeguards in place to help people feel more comfortable with this feature: The audio that is captured is never recorded or stored, and never leaves your device; it is essentially deleted immediately and in real time from local storage; only a truncated version is uploaded to Apple Private Cloud Compute servers for AI processing, and the resulting transcript that is delivered to you is end-to-end encrypted.Still, there are some potential privacy concerns worth considering, especially since you'll essentially be subjecting everyone around you to using it without their consent.How Audio Intelligence makes a transcript of what it hearsSeries 12 and Ultra 4 watches run on an S11 chip with a Secure Exclave feature, a partitioned local storage compartment that even the operating system cannot access. When your Apple Watch picks up ambient audio, it is sent straight into this protected buffer, where it is encrypted, then sent over to a matching Secure Enclave on your paired iPhone. From there, your iPhone transcribes it locally using an on-device AI model to create a summarized version that’s half the length of the actual audio, removing everything but the essentials, and stripping out any identifying information. After the audio data is overwritten, the transcript is sent to Apple's Private Cloud Compute along with some generic metadata, like a rough location and timestamps. None of the audio data is tied to your Apple ID once it reaches the servers at this stage. Apple uses a third-party relay to conceal your IP address while uploading the data to Private Cloud Compute, so not even their employees can identify you from your device. While in the Private Cloud Compute servers, your audio data gets summarized by Siri AI. This is how you end up with daily text summaries of your audio, provided you opt into Siri Recap. Siri Recap summaries are auto-deleted every seven days unless you choose to save them individually. All of this might seem perfectly secure, but here comes the risky part: If you choose to save a conversation summary with Siri Recap, that data is uploaded to your iCloud storage and can even be saved in apps like Apple Notes or Journal. Server-level data breaches at companies like Apple are rare, but account-level hacks happen more often than you might think: 900,000 compromised iCloud logins were found sitting on a public server after a data leak just this January. If someone manages to hack into your Apple ID or access your unsecured iPhone, they might be able to view all your saved conversation summaries from a single place. And it’s not just the potential data breach that should scare you, but all the associated privacy concerns. Series 12 and Ultra 4 watches will now come loaded with Apple's new audio intelligence feature Credit: Apple A potential privacy nightmareApple has spent a lot of time thinking through ways to protect its customers when they use the Audio Intelligence feature, but none of these protections extend to other people who might be sitting across the table or in the same room as you. To its credit, Apple has added an audio chime and a brief visual cue that plays on your watch whenever the Live Rewind feature is triggered. This is intended to inform people around you that they are being listened to. Live Rewind also doesn’t identify speakers, so it won’t name or otherwise identify the person speaking in the transcriptions or summaries—but none of that actually prevents people from being eavesdropped on without their explicit consent. Siri Recap’s implementation is arguably even worse. It has no audio or visual cues to indicate that it’s running, so no one knows that they are being surveilled if you enable this feature. Apple’s privacy documentation is visibly scarce when it comes to the issue of consent, apart from a vague warning on the product pages that advises you to “be mindful of those around you.” This is a potential concern not just for other people who might worry about being spied on every time they see an Apple Watch, but also Apple Watch users themselves, if they are in a region where local laws require consent before recording audio: Some 12 U.S. states have explicit all-party consent laws that forbid undisclosed recordings, while outside the U.S., many E.U. countries have instituted similar requirements; the E.U.'s GDPR also prohibits recordings containing personally identifiable information without disclosure. Using an Apple Watch with the Audio Intelligence feature enabled might sound like it would put you in violation of these laws, but at this point, it seems the device operates in a bit of a grey area—as Apple's privacy documentation notes: Audio Intelligence features do not create audio recordings that can be accessedby the operating system, apps, the user, or Apple. There is no recording to share,forward, or produce if requested by any party, because no recording exists.The Secure Exclave buffer where the audio is initially processed is "a continuously overwritten stream that exists only within the protected hardware and never creates an audio recording," Apple says, meaning no one needs to consent to being recorded. In theory. But that Apple felt the need to go into great detail to explain how the feature technically doesn't violate any privacy laws probably doesn't make it feel any less dystopian to those inclined to view it that way.How to turn Audio Intelligence on or offGiven all the potential security concerns, it’s a good thing that Apple has at least made its Audio Intelligence features opt-in only. They won’t be enabled by default on your new Apple Watch, and you can choose to enable or disable individual features as you choose to. Here's what to look for:Sound Recognition: Open the Settings app on your iPhone. Navigate to Accessibility > Sound & Name Recognition > Sound Recognition. Here you can toggle the “Detect using Apple Watch” option on or off. Music Recognition: Open the Settings app directly on your Apple Watch. Tap on Smart Stack. Switch "Music Detection" on or off per your needs. Live Rewind: Open the Watch app on your iPhone. Go to Siri > Live Rewind. You can choose to enable or disable “Double Click Digital Crown” to control this feature. Siri Recap: Again on the Watch app on your iPhone, go to Siri > Siri Recap. You can enable or disable the feature here, or even set an autorun schedule for it if you like. Siri Recap can also be switched on and off by flipping a single toggle in your Control Center.