When an iPhone asks whether an app may use the microphone and you tap "Don't Allow," you expect that to settle it. For an app already installed on the phone, it didn't always. CVE-2026-65339 was a logic flaw in Apple's audio service that let an app record audio after the user had denied it microphone permission. Two researchers who found the bug published detailed write-ups on Nov. 17.

The good news is that the fix is old. Apple shipped it in iOS and iPadOS 26.6.1 on Aug. 17, and in iOS and iPadOS 18.7.11 for older hardware. Apple also patched its Mac, TV, Vision and Watch platforms. Nobody has reported real-world exploitation, and the flaw isn't on CISA's Known Exploited Vulnerabilities list. The write-ups are still worth reading, though, because they show how a permission prompt that looks absolute can quietly fail.

What Apple said, and what it didn't

Apple's security bulletins are deliberately terse, and this one is too. The bulletin for iOS 26.6.1 and iPadOS 26.6.1 lists CVE-2026-65339 with the impact line "An app may be able to leak sensitive user information" and describes the fix in one sentence: "A logic issue was addressed with improved checks." It credits Mustafa Calap (dbg.re) and Meta Red Team X's Nik Tsytsarkin.

Read on its own, "leak sensitive user information" could mean almost anything. The researchers' disclosure, as reported by The Hacker News, makes it concrete: the sensitive information was audio from the microphone, captured by an app the user had explicitly refused.

A "logic issue" is a bug in reasoning, not in memory handling. Nothing is overflowing a buffer or corrupting memory. Some code path made the wrong decision about whether a request should be allowed. According to The Hacker News, the audio service accepted the app's own claim about which process was asking, and so treated the app as itself and skipped the normal permission check. The fix makes the service verify the real caller's identity, which the kernel supplies.

How the research played out

According to The Hacker News, the researchers tested with sandboxed apps installed through Xcode, Apple's developer toolchain. Those apps never went through App Review, the screening every App Store submission gets. The test apps had no added entitlements, which shows the bug lived in the operating system rather than in some special privilege an app had to be granted. It also means the tests don't show whether App Review would have stopped such an app.

In Mustafa Calap's tests, recording had to start while the app was in the foreground and kept going after it moved to the background or the phone was locked. That's the scenario a microphone permission is supposed to prevent.

iPhones show an orange indicator when an app is using the microphone. The researchers reported that on an earlier test build, the indicator stayed off while recording happened; Tsytsarkin said a later build appears to have restored it. When the permission check and the visual warning both fail, the user has no way to notice.

The Hacker News also noted that neither researcher describes a way for users to check whether an app recorded them through this flaw before it was patched.

How serious is it, really?

The NVD record for CVE-2026-65339 carries a CVSS 3.1 base score of 5.0, which is Medium (the score comes from CISA's entry in the record). The vector string shows why it isn't higher:

  • AV:L (local attack vector): the attacker needs code running on the device. This can't be triggered across the network.
  • AC:L (low complexity): once that code is in place, nothing unusual has to line up for it to work.
  • PR:L (low privileges): an ordinary app is enough. No jailbreak or administrator rights are needed.
  • UI:R (user interaction required): the user has to do something, which here means installing and running the app.

In plain terms, the bug only matters if an untrustworthy app is already on the device. That limits the risk a lot, but it isn't trivial. The entire job of mobile permissions is to protect users from apps they've installed and don't fully trust. A bypass undermines a promise users rely on every time they tap "Don't Allow."

The NVD record also includes CISA's SSVC assessment. SSVC (Stakeholder-Specific Vulnerability Categorization) is a decision framework CISA uses to help organizations prioritize, and it lists exploitation as "none." That agrees with The Hacker News report that neither Apple nor the researchers report any attacks, and with the flaw's absence from the KEV catalog as of that report. KEV is CISA's list of vulnerabilities confirmed to be exploited in the wild, and federal agencies have to patch those on a deadline.

The timeline

The order of events is a little unusual. According to NVD, the CVE was published on Aug. 17, the same day Apple shipped iOS and iPadOS 26.6.1. Apple's bulletin for iOS and iPadOS 18.7.11, the branch for older devices, also lists CVE-2026-65339; that update was released Oct. 26. That bulletin covers iPhone XS, XS Max and XR and the 7th-generation iPad, and it also lists a separate CoreGraphics bug, CVE-2026-86950, which Apple says it is aware may have been exploited in an extremely sophisticated attack against specific targeted individuals on iOS versions before iOS 27. That is a different flaw from the audio bug. The researchers' technical write-ups came out last week, about three months after the first fix.

That gap is normal for coordinated disclosure. Researchers usually hold details until most users have updated. In practice, though, many people heard about the bug only this week, even though the patch has been available since summer.

What It Means for Defenders

Who is affected: Devices running Apple platforms older than the fixed versions below. According to NVD, the flaw is fixed in:

  • iOS 26.6.1 and iPadOS 26.6.1
  • iOS 18.7.11 and iPadOS 18.7.11 (for older devices: iPhone XS, XS Max, XR and the iPad 7th generation)
  • macOS Tahoe 26.6.2
  • macOS Sequoia 15.8
  • tvOS 27, visionOS 27 and watchOS 27

What to do: Install the updates. The researchers describe no setting that blocks the bypass. On any device at or above those versions, the bug is already fixed. If you manage a fleet through MDM, set your minimum OS version to at least these builds and look for devices stuck on older releases, especially older iPhones and iPads on the 18.x branch, which got the fix later than the 26.x branch.

Prioritization: This bug requires an app already on the device. Exploitation isn't reported, it isn't in KEV, and its CVSS score is Medium. It belongs in your normal patch cycle, not an emergency one. That said, the researchers' report that the orange indicator could stay off on an earlier build makes it reasonable to close quickly for executives, staff who handle controlled unclassified information (CUI) on mobile devices, and anyone else whose conversations an adversary would want to hear.

What you can't check: Neither researcher describes a way to tell whether a device was recorded before patching, so don't plan on forensic confirmation from the phone itself. A practical step is to review what's installed on sensitive devices, particularly apps installed outside the App Store through developer or enterprise channels. The researchers' test apps were installed from Xcode and never went through App Review.

The broader lesson: A permission dialog only works if the operating system enforces it in every code path that reaches the resource. Treat "Don't Allow" as strong protection, not a guarantee, and keep untrusted apps off devices that sit in rooms where sensitive conversations happen.

Sources