Security / News explained
ParallelCluster’s log export fix protects the operator’s machine
AWS’s November 9 advisory fixes a path traversal flaw in two ParallelCluster CLI export commands. A writer to the export bucket could affect files outside the intended local directory; version 3.16.0 addresses the issue.
AWS’s November 9, 2026 bulletin for CVE-2026-124033 concerns the AWS ParallelCluster command-line tool, distributed as aws-parallelcluster. Versions from 3.0.0 up to, but excluding, 3.16.0 are affected; 3.16.0 contains the fix. The vulnerable operations are export-cluster-logs and export-image-logs. [1]
The practical consequence sits on the machine running the export command. A diagnostic operation intended to collect logs could write files outside its chosen local directory. For an operator, the security review therefore needs to include the tool that retrieves the evidence.
The export passes through storage
AWS documents both cluster-log and image-builder-log exports as workflows passing through S3. Their command references describe an output-file option for saving the archive locally; otherwise the operation uploads to S3 and returns a URL. The local destination is part of the export workflow, rather than the source of the logs. [3] [4]
Our interpretation is that this workflow has more than one trust boundary. The operator chooses an export and a destination. Storage supplies objects to the retrieval process. The retrieval process then needs to ensure those objects remain inside the location the operator intended.
A hypothetical support engineer might run the export from a workstation, while a diagnostic job might run it from an automation environment. Both uses can look routine because the goal is simply to obtain logs. The review still needs to account for the authority of whichever process performs the local write.
The object key could influence a local path
The project advisory describes an actor who can write to the export S3 bucket and learn the export task ID. By adding a crafted object key during the export, that actor could select a file name and contents outside the export directory, within the command user’s write permissions. [2]
It also says the written file remains after the command exits and that the flaw does not let the actor read files from the export machine. This is a file-write capability; the advisory does not describe unconditional direct code execution. [2]
Those limits matter. A precise exposure assessment asks what the command user could write and what happened during the relevant export. It should preserve the documented capability rather than substitute a broader attack story. At the same time, an unexpected write outside the intended destination is a meaningful boundary failure even when no reading capability is involved.
Consider a hypothetical diagnostic runner with a workspace containing other pending work. A reviewer would want to know whether the export was confined to its designated directory. Merely finding the expected archive would answer whether useful output arrived; it would leave the separate question of unintended output unresolved.
This example does not claim that a runner was altered. It illustrates why an operation’s successful return and its file-system effects are different kinds of evidence. An investigation can examine both without assuming that one proves the other.
Update the tool and review the writers
AWS recommends updating to the corrected release and incorporating the fix into derivative code. Its workaround is limiting access to the log-export bucket to trusted users. [1]
The cluster-export reference also documents that CloudWatch needs permissions for the destination bucket. A permissions review therefore needs to understand the approved logging workflow, rather than treat every writer as interchangeable. [3]
For a team planning the change, our practical recommendation is to identify where the CLI is actually used. The operator’s environment and any automation image that runs exports are relevant places to verify the corrected package. A cluster inventory alone does not establish the version of a separately installed diagnostic tool.
A hypothetical completion record could identify the updated CLI environment, the intended export destination, the approved bucket writers and a successful ordinary diagnostic export. Any review of earlier unexpected files can remain a separate, explicitly owned investigation. These are suggested acceptance criteria, not results from a live cluster test performed for this article.
The broader engineering lesson is that retrieval tools need a clear destination boundary. Logs are useful because they help explain another system. Collecting them should preserve the operator’s chosen scope on the machine doing that work, even when object names arrive from somewhere else.
Sources + further reading
Source dates describe the original material. Publication above is the date this Zyel article went live.
- AWS bulletin 2026-174: ParallelCluster log export path traversalAWS Security · Checked Nov 10, 2026
- ParallelCluster log export security advisoryAWS ParallelCluster project · Checked Nov 10, 2026
- ParallelCluster cluster log export command referenceAWS Documentation · Checked Nov 10, 2026
- ParallelCluster image log export command referenceAWS Documentation · Checked Nov 10, 2026
Reader discussion
What would you ask next?
Questions, useful corrections and thoughtful disagreement are welcome. Comments are reviewed before they appear.
No reader comments yet.