iPhone flaw bypassed denied microphone permission
An iPhone flaw in Apple’s audio service let installed apps record audio after users denied microphone permission. The issue, tracked as CVE-2026-65339, is now patched. Apple’s first security updates containing the fix were released on August 17, and neither Apple nor the researchers who reported the flaw said it had been used in attacks.
The technical details were published by Meta Red Team X, while a security engineer separately released an analysis of the same issue. In testing, the microphone permission prompt appeared when an app was opened for the first time. Tapping “Don’t Allow” did not stop recording, and the prompt did not return during later launches.
On iPhones, an orange indicator normally shows when an app is using the microphone. In an earlier test build, that indicator stayed off while the app recorded. The researcher’s demo video shows the indicator during recording without microphone permission, but the video was made on a later build that appeared to have restored the orange indicator. The separate analysis reported recording while permission still showed as denied, with no orange indicator in the Dynamic Island or Control Center. It did not identify which of the two test builds produced that result.
The researchers did not describe any way for users to determine whether an app had recorded audio before the fix. The bypass was reproduced on iOS 26.6, released July 27, and on an iPadOS 27 beta. It required an app already installed on the device, but the test apps were sandboxed apps, meaning they were restricted from freely accessing the rest of the system, and they had no extra entitlements, which are special permissions granted to an app. Their only privacy declaration was NSMicrophoneUsageDescription, the setting that causes the microphone permission prompt to appear.
Recording had to begin while the app was in the foreground, meaning visible and active on the screen. After it started, it continued when the app moved into the background and when the device was locked. The background test also used UIBackgroundModes = audio. Both test apps were installed from Xcode and never passed through Apple’s App Review, so the testing does not establish whether App Review would have rejected such an app.
The Meta analysis compared two iOS 27 test builds. Neither researcher reported testing macOS, tvOS, watchOS or visionOS, although Apple listed fixes for those systems as well. The bypass affected microphone capture through the audio service only. It did not provide access to other privacy permissions.
Apple released iOS 26.6.1 and iPadOS 26.6.1 on August 17 for iPhone 11 and later, iPad Pro 12.9-inch 3rd generation and later, iPad Pro 11-inch 1st generation and later, iPad Air 3rd generation and later, iPad 8th generation and later, and iPad mini 5th generation and later. For older supported devices, iOS 18.7.11 and iPadOS 18.7.11 arrived on October 26 for iPhone XS, iPhone XS Max, iPhone XR and iPad 7th generation.
The other listed fixes are macOS Tahoe 26.6.2 on August 17, macOS Sequoia 15.8 on September 14, tvOS 27 on September 14 for Apple TV 4K 2nd generation and later, watchOS 27 on September 14 for Apple Watch Series 9 and later, and visionOS 27 on September 14 for Apple Vision Pro models. Meta’s timeline says the fix shipped in iOS 27.0 beta 5 on August 10. The iOS 27 release notes did not list the flaw, nor did the notes for later iOS 27 updates as of the latest information available.
The researchers described no device setting that blocks the bypass, making installation of the relevant updates the stated mitigation. CISA assigned the flaw a CVSS score of 5.0 out of 10. CVSS is a scale used to describe the severity of a security vulnerability. CISA’s catalog of known exploited vulnerabilities did not list CVE-2026-65339 at the time of its update late the previous week.
Apple’s August 17 security notes described the issue only as one that might let an app leak sensitive user information. The Meta researcher said that wording did not spell out the most significant behavior, namely recording after permission had been denied without the orange indicator appearing. In the researcher’s own tests, however, the missing indicator occurred only on the earlier build.
The technical cause involved Apple’s Transparency, Consent, and Control system, or TCC, which decides whether a process is allowed to access protected resources such as the microphone. The audio service, called audiomxd, accepted an app-provided audit token. An audit token is information used to identify the process making a request. An app could provide a token identifying audiomxd itself, causing the service to treat the app as its own internal process and use a shortcut intended for internal audio operations. The kernel, the core part of the operating system, supplied the information the app needed to impersonate the service.
The fix makes the audio service verify the real caller identity supplied by the kernel. An outside caller now needs the private entitlement com.apple.private.audio.client-audit-token-override to replace the client identity used by the service, and an app cannot grant itself that entitlement. For businesses and IT teams managing Apple devices, the practical lesson is to track the listed operating-system updates rather than rely only on microphone permission settings. AEU-I provides security-first IT, infrastructure and consulting, which can be relevant when organisations need help managing security responsibilities across their IT environment.