# Apple audiomxd Audit Token Forgery Lets Sandboxed Apps Record Microphone Audio Without Permission (CVE-2026-65339)

> A logic flaw in Apple's audiomxd audio media daemon trusted a client-supplied audit token that could be forged, letting a sandboxed app already installed on the device capture microphone audio without a TCC microphone grant. Apple fixed it in iOS/iPadOS 26.6.1, macOS Tahoe 26.6.2 and later releases; no in-the-wild exploitation is reported.

- **Published:** 2026-11-16T10:52:38Z
- **Last reviewed:** 2026-11-16T11:49:01.589Z
- **Canonical:** https://intel.threadlinqs.com/threat/TL-2026-6133
- **ID:** TL-2026-6133
- **Severity:** MEDIUM (CVSS 5)
- **Category:** VULNERABILITY
- **Status:** PATCHED
- **Detections:** 9 (1 shown free below) · **IOCs:** 15
- **CVEs:** CVE-2026-65339

## Description

CVE-2026-65339 is a logic/authorization flaw in audiomxd, Apple's audio media daemon, which runs the server side of AURemoteIO (the low-latency recording audio unit). Property 2009 ("client audit token") in AUIOServer_SetProperty, reachable through MIG messages, copied a client-supplied 32-byte value into the stored client audit token and then called AQIONodeClient::SetCanRecord(). The handler never checked the caller's kernel-supplied audit token.

SetCanRecord() compared the stored token's PID and pidversion against the daemon's own. On a match it treated the client as hosted inside audiomxd and allowed recording without consulting TCC (the log message says recording was "implicitly allowed" for a client "hosted in audiomxd"). The TCC mismatch path also used the same client-supplied token as its subject, so attacker-controlled identity influenced both branches.

According to the reporting, a sandboxed app cannot look up the daemon's PID directly, but the researchers obtained it from the kernel-attached trailer of an error reply to a malformed request, then forged the token through the public AudioUnitSetProperty call using property 2009. Guessing the process version also worked, taking about 70 seconds in testing. Impact: microphone capture without a microphone grant or special entitlements; recordings were attributed to the daemon rather than the app in logs, and the orange microphone indicator did not appear on the analyzed beta build. The first-launch prompt still appeared, but declining did not stop recording and the prompt did not return on later launches. The bypass was limited to microphone capture through this audio path and did not grant general TCC permissions. It is not remotely exploitable: a malicious app must first be installed.

The fix (build 24A5418b of iOS 27 beta 5) makes the property 2009 handler check the caller's kernel-supplied audit token before copying the client token. Identity changes are allowed only if the caller presents the sentinel token used by the daemon's in-process clients or holds the private entitlement com.apple.private.audio.client-audit-token-override. Apple describes it as "a logic issue was addressed with improved checks", impact "an app may be able to leak sensitive user information". Reported to Apple on 2026-07-07 by Mustafa Calap and Nik Tsytsarkin (Meta Red Team X). The primary write-up was published 2026-11-09. No in-the-wild exploitation, malware family or network IOC is reported in the sources; CISA KEV inclusion could not be confirmed from the feed (the large file was only partially readable) and no source claims it.

## MITRE ATT&CK

- T1123 Audio Capture
- T1429 Audio Capture
- T1548 Abuse Elevation Control Mechanism
- T1036 Masquerading
- T1685 Disable or Modify Tools
- T1057 Process Discovery
- T1404 Exploitation for Privilege Escalation
- T1068 Exploitation for Privilege Escalation
- T1655 Masquerading
- T1629 Impair Defenses

## Free detection (Splunk SPL, Microsoft KQL, Sigma)

**CVE-2026-65339 audiomxd Implicit Microphone Allow for Non-Apple Client — Exploitation** (`TL-2026-6133-SPL-001`)

Flags the audiomxd log path for CVE-2026-65339 (audit token forgery via AudioUnitSetProperty property 2009 / AQIONodeClient::SetCanRecord) where recording is 'implicitly allowed' for a client 'hosted in audiomxd' without a TCC microphone check, when the message does not reference an Apple-signed client. On an unpatched device this is the direct trace of a sandboxed app forging the daemon's own identity to record microphone audio without a grant.

- **Languages:** Splunk SPL, Microsoft KQL, Sigma
- **ATT&CK:** T1123, T1429, T1548
- **Data sources:** macos_unified_log, splunk index=macos sourcetype=macos:unifiedlog:json
- **Severity:** medium
- **Confidence:** medium

### Splunk SPL

````spl
index=macos sourcetype="macos:unifiedlog:json" earliest=-24h processImagePath="*/audiomxd" (eventMessage="*implicitly allowed*" OR eventMessage="*hosted in audiomxd*") NOT eventMessage="*com.apple.*" ```CVE-2026-65339 audiomxd SetCanRecord bypass: implicit allow for a non-Apple client``` | stats count as allow_events min(_time) as first_seen max(_time) as last_seen values(eventMessage) as sample_messages by host | convert ctime(first_seen) ctime(last_seen) | table host allow_events first_seen last_seen sample_messages
````

### Microsoft KQL

```kql
MacUnifiedLog_CL
| where TimeGenerated > ago(24h)
| where ProcessImagePath endswith '/audiomxd'
| where EventMessage has_any ('implicitly allowed', 'hosted in audiomxd')
| where EventMessage !has 'com.apple.'
| summarize AllowEvents = count(), FirstSeen = min(TimeGenerated), LastSeen = max(TimeGenerated), Samples = make_set(EventMessage, 5) by Computer
```

### Sigma

```yaml
title: Apple audiomxd Implicit Microphone Allow for Non-Apple Client (CVE-2026-65339)
id: 6f1c2a54-3b7e-4d1a-9c58-0a1e7b2d4f61
status: experimental
description: Detects the audiomxd unified-log message indicating recording was implicitly allowed for a client hosted in audiomxd, without an Apple client reference. Indicates the CVE-2026-65339 audit token forgery path on an unpatched device.
references:
  - https://rtx.meta.security/exploitation/2026/11/09/CVE-2026-65339.html
  - https://nvd.nist.gov/vuln/detail/CVE-2026-65339
author: Threadlinqs Team
date: 2026-11-16
tags:
  - attack.collection
  - attack.t1123
  - attack.t1429
  - attack.privilege-escalation
  - attack.t1548
  - cve.2026.65339
logsource:
  product: macos
  service: unifiedlog
detection:
  selection:
    processImagePath|endswith: '/audiomxd'
    eventMessage|contains:
      - 'implicitly allowed'
      - 'hosted in audiomxd'
  filter_apple:
    eventMessage|contains: 'com.apple.'
  condition: selection and not filter_apple
falsepositives:
  - Legitimate in-process audio clients hosted in audiomxd that are not labelled com.apple.
  - Patched hosts where sentinel-token in-process clients still log the same message
level: medium
```

**Known false positives:**

- Legitimate in-process audio clients hosted in audiomxd whose log line does not contain com.apple.
- Patched hosts where sentinel-token in-process clients still emit the same message

**Tuning notes:** The index, sourcetype and field names (processImagePath, eventMessage) assume macOS unified log forwarded as JSON (log show --style ndjson or a Jamf Protect/Splunk forwarder); adjust to your pipeline. Message wording comes from the public write-up, so confirm it on a lab Mac before alerting. Baseline 7 days on patched hosts and allowlist known in-process clients. Visibility is macOS only; iOS/iPadOS devices do not expose unified logs to a SIEM, so use MDM patch compliance (SIG-001) there.

### 8 more detection rules (Splunk SPL, Microsoft KQL, Sigma), unlock with TL Creds

- Private audio client-audit-token-override Entitlement Referenced Outside [indicator] — Exploitation (locked: unlock with TL Creds)
- Microphone Prompt Declined Yet [indicator] Allows Recording — Post-Exploitation (locked: unlock with TL Creds)
- [indicator] Rejections of Client Audit Token Override (pidversion Brute Force) — Reconnaissance (locked: unlock with TL Creds)
- Apple Device Running OS Vulnerable to CVE-2026-65339 [indicator] Flaw — Exposure (locked: unlock with TL Creds)
- Repeated [indicator] Implicit Recording Allows on One Host — Sustained Eavesdropping (locked: unlock with TL Creds)
- Repeated [indicator] Service Exits — Failed Exploitation Attempts (locked: unlock with TL Creds)
- [indicator] Recording Allowed Without Any TCC Microphone Event — Post-Exploitation (locked: unlock with TL Creds)
- [indicator] MIG Error-Reply Burst (Daemon PID Leak Probe) — Reconnaissance (locked: unlock with TL Creds)

Sign up free to unlock the detection rules: https://threadlinqs.com/?modal=signup

## Sources

- [Apple Fixes iOS Flaw That Let Sandboxed Apps Secretly Record Microphone Audio Without Permission](https://cybersecuritynews.com/apple-fixes-ios-flaw/)
- [Meta Red Team X: CVE-2026-65339 audiomxd write-up](https://rtx.meta.security/exploitation/2026/11/09/CVE-2026-65339.html)
- [NVD: CVE-2026-65339](https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-65339)
- [Apple security release notes (148282)](https://support.apple.com/en-us/148282)
- [Apple security release notes (148281)](https://support.apple.com/en-us/148281)
- [Apple security release notes (149225)](https://support.apple.com/en-us/149225)
- [Cyber Security News: Apple fixes 28 security vulnerabilities](https://cybersecuritynews.com/apple-fixes-28-security-vulnerabilities/)

## Full data

The other 8 detection rules (Splunk SPL / Microsoft KQL / Sigma) unlock with TL Creds; a free account includes daily creds: https://threadlinqs.com/?modal=signup
Indicator values unlock with TL Creds, at a lower price on Red and Purple; AI agents can query rules and IOCs through the Threadlinqs MCP server (Purple): https://intel.threadlinqs.com/mcp

Canonical: https://intel.threadlinqs.com/threat/TL-2026-6133
