Key points
- CVE-2026-14904 allows authenticated remote users to read any file readable by root on cluster-manager instances.
- The flaw exists in the Auth.GetUserPrivateKey API due to improper link resolution before file access.
- AWS released RES version 2026.06 to fix the issue and provided patch scripts for previous major versions.
Authenticated users of AWS Research and Engineering Studio could read arbitrary files on cluster-manager EC2 instances by exploiting a symbolic link vulnerability in the authentication API. The issue, tracked as CVE-2026-14904, was addressed in RES version 2026.06 released on 7 July 2026, according to AWS Security Bulletins.
The vulnerability arises from improper link resolution in the Auth.GetUserPrivateKey function. An attacker could replace their SSH private key file with a symbolic link pointing to any host file. Because the cluster-manager process runs as root, this grants access to all files readable by the root user, including other users' SSH keys and application secrets.
Scope and Impact
AWS Research and Engineering Studio is an open-source solution for managing secure virtual desktops and computing resources on AWS. The bulletin rates the issue as important, requiring attention. The exposure affects the cluster-manager EC2 instance specifically, where the process privileges allow broad file access.
Who is Affected
Users running versions of RES prior to 2026.06 are affected. This includes any forked or derivative code that has not incorporated the new fixes. AWS recommends upgrading to the latest version immediately. Customers unable to upgrade right away can use patch scripts available for the past three major versions, with instructions on the RES GitHub wiki.
What happens next
Teams should verify their RES version and upgrade to 2026.06 if they are running an older release. For those maintaining forked code, applying the specific patches from the GitHub wiki is necessary. AWS advises contacting [email protected] for any security concerns related to this bulletin.
Why this attack works
This technique relies on the application trusting user-supplied file paths without validating the underlying filesystem objects. When a service processes a symbolic link as if it were the target file, it bypasses intended access controls. In intrusion scenarios, this serves as a privilege escalation or data exfiltration vector, allowing attackers to read sensitive configuration data or credentials that the application process can access but the user should not.
What to do and how to stay safe: AWS
- Check your AWS Research and Engineering Studio version against the 2026.06 release notes.
- Review symbolic link permissions on cluster-manager EC2 instances to limit potential abuse.
- Monitor logs for unusual access patterns to the Auth.GetUserPrivateKey API endpoint.
- Apply the available patch scripts if an immediate upgrade to the latest version is not feasible.
Step-by-step guide: Patch Management Policy: How to Automate Fixes Without Breaking Systems
General security guidance from the Payload Report newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
What is CVE-2026-14904?
It is an improper link resolution vulnerability in the Auth.GetUserPrivateKey API of AWS Research and Engineering Studio.
Can unauthenticated users exploit this flaw?
No, the advisory states that an authenticated remote user is required to perform the attack.
Is a patch available for older versions?
Yes, patch scripts are available for the past three major versions, while a full fix is in version 2026.06.
