macOS Tahoe 26.7 contains a previously overlooked framework that controls the image capture of the upcoming AirPods. It describes two cameras, two recording modes, and a light intended to indicate to others that a recording is taking place. However, at one crucial point, the code falls short of expectations.
The feature was revealed earlier this week via a video included in the macOS Tahoe 26.7 release candidate: In it, a person holds a book cover up to their earbuds to memorize the information. The demo video, drawn from Apple's own code, showcases the feature in a short sequence. However, the same system version contains much more than just the video and translation table.
Forum member mactracker analyzed the remaining data and documented the results in the MacRumors forum. Each statement there is supported by a source that can be verified via the public system comparisons of the ipsw-diffs project on GitHub.
Key Facts at a Glance
- A framework called AccessorySensorManager controls image capture when the earbuds are connected to a Mac.
- Each earbud contains a color camera and delivers synchronized image pairs.
- There is an active and a passive mode with different resolutions; video is not included.
- The code controls an indicator light and its brightness, which should make the recording visible.
- Person recognition is present, but it explicitly does not suppress recordings in the code.
Two cameras, one image pair
According to the findings, each earbud contains its own image sensor. Apple refers to the hardware as "Kino" and distinguishes between a left and a right camera for machine vision in the code.
The two devices work as a pair. Each recording is assigned an image identifier, and the system only considers a recording complete when the identifiers of both sides match – if they differ, macOS interprets this as an error. If only one earbud delivers an image, the system can create an empty placeholder for the missing side.
The color images are output in a format with three separate 16-bit color layers. This confirms that they are color images and not grayscale or pure infrared data. However, the discovery does not rule out the possibility of additional sensors operating outside this image sequence.
Active and passive: two resolutions
The framework has two operating modes that differ significantly in resolution and purpose.
| Mode | Recording | Output |
|---|---|---|
| Active | 640 × 640 pixels | 1024 × 1024 pixels |
| Passive | 320 × 320 pixels | 320 × 320 or 512 × 512 pixels |
The active mode thus corresponds to approximately 0.4 megapixels of captured image data and just over one megapixel in the output image. These are values for the respective shooting distance – the physical resolution of the sensor may be higher.
Video recording is not supported. The framework supports single images, either as a one-time capture or repeatedly at a set rate. This aligns with the functionality of Visual Intelligence, which is designed to understand individual images – on the iPhone, Visual Intelligence uses the camera and screen according to the same principle.
What could trigger passive mode
The low-resolution mode apparently serves to perceive the environment. The system recognizes five conditions for this: speech in the vicinity, a change in the acoustic environment, a change in body posture, a head turn, and leaving a defined radius.
The code does not indicate whether each of these conditions actually triggers a recording. It also remains unclear what type of language is recognized.
The warning light and its gap
From the perspective of bystanders, the most important finding is a hardware display whose status and brightness can be controlled from within the system. This suggests a light that indicates an ongoing recording – similar to the green camera indicator on an iPhone.
The crucial point is what the code doesn't address here. There's no indication that an unobstructed view of this light is a prerequisite for recording. Therefore, a light obscured by hair or a hat wouldn't prevent recording.
The situation is even clearer with regard to person recognition. The earbuds themselves run an analysis that determines whether a person is in the field of view. One might naturally assume this provides a protective function. However, the code points in the opposite direction: it registers a result indicating an unrecognized person and yet still transmits the image. No mechanism that prevents recordings of unidentified individuals is found in the data.
Calibration against constant head movement
Because earbuds sit in the ear, are oriented differently, and constantly move while walking, the framework invests considerable effort in calibration. It takes into account information about the field of view and optical center, the position of the cameras relative to each other and to the user, lens distortion correction, and a comparison between the camera and the motion sensor.
Distortion correction is the most revealing aspect: it compensates for the curvature produced by particularly small or wide-angle lenses. The system can also filter out images with excessive motion blur or obstructed views.
Why the findings likely describe a cancelled model
The most important caveat lies in the codename. The earbuds shown in the video are codenamed B790, a camera variant of the AirPods Pro 3. According to Mark Gurman, this exact model will no longer be released; instead, Apple is working on a new generation codenamed B798, which will still be launched in 2027.
The findings likely describe how Apple's camera earbuds could work – not what features the device will ultimately have.
The confusing nature of the situation is illustrated by the developments of the past few weeks. At the beginning of August, it was still being said that a second camera variant could be released this year; at the beginning of July, a leaker even reported a development halt. Both of these statements remain alongside the most recent assessment and can be found side-by-side in the overview of the AirPods Ultra.
The light alone will not solve the problem
A visible indicator is the right approach, and the fact that it's even in the code suggests that Apple is taking the issue seriously. However, it's less effective on earbuds than on an iPhone. Lifting a phone already signals that a recording is being made; an earbud sits discreetly on the head, often partially covered by hair, and isn't perceived as a camera by those around you.
Then there's the passive mode. A recording triggered by a head turn or a sound occurs without the wearer's conscious action – they themselves aren't sure when their device is taking a picture. In Germany, Austria, and Switzerland, this raises a question for you as the wearer that the technology doesn't answer: Anyone who photographs other people in public spaces is responsible for it, even if it was triggered automatically.
That's precisely why the finding regarding person recognition is so significant. The function exists, it recognizes people in the field of view – and according to the evidence, the code does not use this information to withhold recordings.
A draft version, not a data sheet
These figures are likely to change significantly before the official launch, especially since the data comes from a model that isn't intended for release. As Apple indicates, this is an open-ended test against which the earbuds will ultimately be measured.
Would a light on the earbud be enough for you to feel comfortable around a stranger wearing it – or do you need a blocker that prevents recordings with other people from the outset? Let us know in the comments what solution you expect.
- Apple Music labels AI-generated songs with "Made With AI"
- Stick: Season 2 starts on November 4th.
- Apple rejects OpenAI's dismissal request
