macOS 26 Tahoe and system audio — what changed, and what didn’t

macOS 26 Tahoe did not add system-audio recording to Apple’s own tools. The screenshot toolbar and QuickTime are still microphone-only, and the change worth knowing about is somewhere else entirely: Apple now documents a process-level audio tap, which lets an app capture what your Mac is playing without you installing a virtual audio driver first.

That distinction is why so many pages on this topic are wrong. A capability got documented in the macOS 26 docs, several blogs read that as new in macOS 26, and a few went further and told readers the built-in recorder can now do it. Apple’s own user guide contradicts them in one sentence.

What Apple’s recorder actually offers on Tahoe

Open the screenshot toolbar with ⌘⇧5, click Options, and look at the audio section. Apple’s macOS Tahoe 26 guide describes the whole of it:

Below Microphone, select the microphone you want to use so that a checkmark appears next to it. To turn off the microphone, select None.

A microphone picker and an off switch. No system audio, no internal audio, no “include computer sound” — the words do not appear on the page. QuickTime Player’s audio recorder is the same story, which is its own page.

So if you have been recording calls with ⌘⇧5 and getting only your own half, nothing on your Mac is misconfigured. You used a tool that was never built to capture the other side.

The permission panel, and why it says “Screen”

The control you want is in System Settings > Privacy & Security > Screen & System Audio Recording. Apple’s support page for it — titled Control access to screen and system audio recording on Mac — states the split plainly:

You can allow apps to record both your screen and audio, or just your audio.

Screen and audio are separable inside one privacy control, which means an app that records audio and no pixels has no business asking for screen access, and you should be suspicious of one that does. It also explains the mildly confusing result: an audio-only recorder shows up in a list whose name begins with the word Screen.

What Apple does not tell you is when the panel got that name. The support page carries the current title while covering macOS Tahoe 26, Sequoia 15, Sonoma 14, Ventura 13, Monterey 12, Big Sur 11 and Catalina 10.15 all at once, so it dates nothing. Third-party write-ups variously credit Sonoma and Sequoia with the rename. I could not find an Apple source for either, so I am not going to pick one. The practical version: if an instruction tells you to open “Screen Recording”, the entry you are looking for now reads “Screen & System Audio Recording”, and a grant carried over from an older macOS is worth re-checking under the new entry.

The change that matters: a process-level audio tap

Apple’s developer documentation now carries an article called Capturing system audio with Core Audio taps, and its one-line abstract is the whole idea:

Use a Core Audio tap to capture outgoing audio from a process or group of processes.

Two details from that article are the ones a user cares about. First, the permission:

To capture audio with a tap, you need to include the NSAudioCaptureUsageDescription key in your Info.plist file, along with a message that tells the user why the app is requesting access to capture audio.

Second, when you find out about it:

The first time you start recording from an aggregate device that contains a tap, the system prompts you to grant the app system audio recording permission.

That is a distinct prompt from the microphone one, and distinct again from screen recording. And CATapDescription, the object that configures a tap, is described as “a tap object that contains an input stream. The input stream is a mix of all of the specified processes output audio.”

Now the version arithmetic, because this is exactly where the confident wrong pages go wrong. The article’s availability is macOS 26.0. The function it teaches, AudioHardwareCreateProcessTap, is available from macOS 14.2. The capability predates Tahoe by two major releases; the documentation is Tahoe-era.

Worth knowing about taps in general, from the same article: a tap can be public (visible to all users) or private to the process that created it, and a tap can mute the process it is capturing so its output stops reaching your speakers and goes only to the tap. If a recorder ever silences your call, that is the mechanism.

Three routes, and what each one costs

One honesty note on the first row: I am describing what the driver is — macOS presents it as an audio input device — rather than claiming which TCC entry your recorder ends up in. I could not find an Apple page that states that, so I am not stating it either.

The middle row is the one to watch if you only want audio. ScreenCaptureKit’s own framework page tells developers to add NSScreenCaptureUsageDescription and request screen-recording permission before capturing anything. So an app built on ScreenCaptureKit asks for your screen even when it only wants the sound, because that is the grant the framework runs on. An app built on a Core Audio tap does not.

What is genuinely broken right now

Three things, each with a source and a date, because “is this a bug or me” is the real question when a recording comes out empty.

All-zero audio buffers on macOS 15 Sequoia. A developer reported that a HAL-plugin capture path went silent on Sequoia while working on Sonoma and earlier: “with the latest macOS 15 Sequoia, all audio buffer values are zero.” The workaround they found was ordering — set the default audio device after starting the session rather than before. Filed as FB15620713 in October 2024, with an Apple engineer asking follow-up questions on the thread and no resolution posted. Nothing you can fix in Settings; it is a fix the app has to ship.

Plain executables missing from the permission list on macOS 26.1. Non-bundled command-line binaries prompt for screen recording, work once granted, and then never appear under Screen & System Audio Recording — so you cannot revoke them from the UI. Proper .app bundles are listed normally. Apple’s Developer Technical Support answered on the forums in November 2025: “IMO this is a bug and I encourage you to file it as such.” If you are a normal person using normal apps, this one will never touch you.

Left channel only from multichannel USB interfaces. An Apple Community thread reports that native screen recording and QuickTime capture only the left channel — input 1 — from interfaces including the Focusrite Scarlett 8i6 and 4i4, the Vocaster One, and the Apogee Symphony Desktop, on macOS 26.0.1 and 26.1. OBS captures correct stereo from the same hardware, which points at Apple’s capture path rather than the drivers. The most recent confirmation in that thread is November 2025; I found no Apple acknowledgment, and no report either way for the later 26.x point releases, so do not read this as a live status.

What I could not verify, stated as such

• Which macOS release renamed the panel. Not documented by Apple as far as I could find.
• Whether the left-channel defect survives on current 26.x. The last data point in the thread is nine months old.
• **Whether ScreenCaptureKit’s audio capture needs anything beyond the screen-recording grant.** Apple’s framework page names NSScreenCaptureUsageDescription for capture generally and does not separately address the audio case. Third-party blogs assert it does not; that is not a source.
• Any claim that Tahoe's built-in recorder gained system audio. Directly contradicted by the guide quoted above. If a page tells you Tahoe records internal audio out of the box, it did not check.

Where Narada sits in this

Narada takes the third row of that table. System audio comes from a Core Audio process tap, so there is no virtual device to install, nothing to undo in Audio MIDI Setup, and no Screen Recording permission — it asks for Microphone and System Audio Recording, and that is the list. Minimum macOS is 15.6.

The microphone and the system output are captured as two separate streams and transcribed separately, on-device, using Parakeet TDT v3 across 25 European languages. Accuracy is not one number: the model’s published word error rates run from about 4% on Italian to about 37% on Greek. The call’s audio gets unnamed speaker labels — Speaker 1, Speaker 2 — because nothing on your Mac knows who those people are; a microphone-only recording gets none, since the labels come from separating voices in the call track.

And given that this whole page is a list of ways macOS audio capture fails: Narada checks each finished recording against itself, comparing the frames it submitted against the frames actually in the file, and marks the recording Incomplete with a reason when audio went missing. It cannot get the lost audio back. It can stop a half-recorded meeting from looking like a whole one, which is more than the silence you get from the tools above.

None of the permission plumbing on this page has anything to do with whether you are allowed to record. Recording rules vary by country and by state, no app can grant you permission, and nothing here notifies the other participants on your behalf — that sentence at the start of the call is yours to say.

Sources

• Apple Support, macOS User Guide (Tahoe 26 edition), Take a screenshot or screen recording on Mac — the microphone-only audio option — https://support.apple.com/guide/mac-help/take-a-screenshot-or-screen-recording-mh26782/mac
• Apple Support, Control access to screen and system audio recording on Mac — the panel name and “both your screen and audio, or just your audio” — https://support.apple.com/guide/mac-help/control-access-screen-system-audio-recording-mchld6aa7d23/mac
• Apple Developer, Capturing system audio with Core Audio taps — abstract, NSAudioCaptureUsageDescription, the system-audio prompt, tap muting and visibility; article availability macOS 26.0 — https://developer.apple.com/documentation/coreaudio/capturing-system-audio-with-core-audio-taps
• Apple Developer, AudioHardwareCreateProcessTap — availability macOS 14.2 — https://developer.apple.com/documentation/coreaudio/audiohardwarecreateprocesstap(_:_:)
• Apple Developer, CATapDescription — “a mix of all of the specified processes output audio” — https://developer.apple.com/documentation/coreaudio/catapdescription
• Apple Developer, ScreenCaptureKit framework — NSScreenCaptureUsageDescription and the screen-recording request — https://developer.apple.com/documentation/screencapturekit
• Apple Developer, SCStreamConfiguration.capturesAudio (macOS 13.0) and captureMicrophone (macOS 15.0) — https://developer.apple.com/documentation/screencapturekit/scstreamconfiguration/capturesaudio
• Apple Developer Forums thread 767333 — Sequoia all-zero audio buffers, FB15620713, the startRunning ordering workaround — https://developer.apple.com/forums/thread/767333
• Apple Developer Forums thread 807898 — plain executables absent from the Screen & System Audio Recording list on 26.1, and the DTS reply — https://developer.apple.com/forums/thread/807898
• Apple Community thread 256163770 — left-channel-only capture with multichannel interfaces on 26.0.1 and 26.1 — https://discussions.apple.com/thread/256163770
• Apple Developer, macOS landing page — currently headlines “What’s new in macOS 27” — https://developer.apple.com/macos/

Every URL above was fetched on 4 August 2026.