What Apple says, what the PoC shows, and where CVE-2026-84568 evidence stops
A source-by-source review of the CVE-2026-84568 vendor impact claim, public reproduction and remaining evidence boundaries.
ecurity reporting often compresses a vendor advisory and independent research into one sentence. CVE-2026-84568 shows why that habit can mislead. Apple describes a possible root code-execution impact. The public reproduction demonstrates a mount trust-boundary failure and explicitly stops short of proving code execution on its tested paths.
Apple fixed the issue in macOS Sequoia 15.8, macOS Tahoe 26.7 and macOS Golden Gate 27. The weakness is CWE-22 path traversal. NVD records CVSS 3.1 score 7.8 with a local attack vector, low complexity and low privileges.
The vendor statement
Apple's impact description says an attacker with control of a network directory server may be able to execute arbitrary code with root privileges. That is a consequential statement because automountd operates in a privileged context and consumes directory-provided mount maps.
The fixed releases establish that Apple considered the issue worth a security update on three named macOS branches. The advisory does not turn every Mac into an exposed host. A client still needs the relevant directory-backed automount configuration and a record that causes the vulnerable behavior.
The public test
The independent repository reverse-engineers the patched function and builds a test around LDAP maps and a local NFS export. Its result is that a vulnerable automountd accepts a map source resolving to the local machine, mounts the export at an attacker-selected path and exposes a marker file through that mount.
The report names the patched function sym.func.100005bd8. It says the patch rejects entries matching localhost, the local hostname or a local IP address. The tested mount source is 127.0.0.1.
That evidence supports a clear claim: the client had a trust-boundary defect around local-source mount records. It also supports investigation of directory changes and local-source mounts on vulnerable clients.
The evidence boundary
The repository says its PoC did not demonstrate arbitrary code execution, privilege escalation or a shell on the victim. It examined an unrelated fstype injection idea and reported that stock macOS path resolution blocked the tested path. It labels that as separate research, not as part of CVE-2026-84568.
These statements do not refute Apple. They define what the public artifact proves. Apple remains responsible for the advisory's root impact claim. The independent researcher remains responsible for the reproduction's narrower result.
A useful incident report can preserve both statements without implying that one validates the other. For example: Apple rates the issue as a possible root code-execution flaw; the public reproduction confirms an accepted local-host automount source but did not achieve code execution on its tested paths. That sentence keeps the source and evidence boundary intact.
What to verify internally
Inventory Macs that query LDAP, NIS or Open Directory for automount maps. Confirm their effective macOS build against the fixed releases. Review directory records for local-host sources and check who can modify the relevant containers.
If a suspicious mount is found, collect the map, mount path, source address, timestamp, endpoint hostname and associated directory change. Determine whether the mount was successful. A failed mount can still show that automountd accepted the source, but it does not prove code execution.
Look for signs of post-mount execution separately. Apple's impact statement makes that a legitimate investigation question, but the public PoC should not be cited as proof that it happened. Preserve process, authentication and file-integrity evidence before remediation removes context.
Response and assurance
Patch to Sequoia 15.8, Tahoe 26.7, Golden Gate 27 or later. Restrict automount record changes to the smallest administrative group. Review recent changes and local-source records across every client that consumes the same directory data.
After remediation, verify that endpoints run a fixed build and that the directory contains only approved records. Add a detection case for a directory-backed mount whose source resolves to the endpoint itself. Distinguish that event from routine local mounts created by unrelated software.
A ZoomEye query for app="macOS" returned 1,199 product-fingerprint matches on November 15, 2026. It does not resolve the evidence boundary: the result identifies a broad macOS fingerprint population, not clients with vulnerable autofs records or confirmed exploitation.
References
Did you enjoy this article?
Recommend it — Standard Reader surfaces well-loved writing to more readers across the network.