CVE-2026-14904: AWS RES Symbolic Link Vulnerability Allows Arbitrary File Read as Root

Technical summary

CVE-2026-14904 is an insecure symbolic link resolution vulnerability (CWE-59) in the Auth.GetUserPrivateKey API endpoint of AWS RES. The vulnerability exists because the system fails to validate or dereference symlinks before reading the file at ~/.ssh/id_rsa during the private key retrieval operation. An authenticated remote attacker can craft a symlink at their SSH key location that points to any file on the cluster-manager EC2 instance. Since the cluster-manager daemon executes as root, the attacker can read arbitrary files accessible by the root user, including SSH private keys from other accounts, database credentials, API tokens, and other sensitive configuration data. The vulnerability affects the authentication workflow and represents a path-traversal variant that operates through filesystem link manipulation rather than directory traversal sequences.

Business impact

This vulnerability enables lateral movement and credential theft within AWS RES deployments. An attacker who gains initial user access can exfiltrate SSH keys and secrets belonging to other users or administrative accounts, potentially compromising the entire cluster and downstream AWS infrastructure. In multi-tenant or shared research environments, this creates significant data exposure risk. The ability to read arbitrary files as root gives attackers access to application configuration, database connection strings, and tokens that may extend compromise beyond the RES cluster to databases, registries, and cloud APIs. Organizations using RES for sensitive research or engineering workloads face confidentiality breaches that could expose intellectual property, credentials, or research data.

Affected systems

AWS Research and Engineering Studio versions prior to 2026.06 are affected. The vulnerability applies to any RES deployment where the cluster-manager EC2 instance is running an affected version. Exposure is limited to users with valid RES authentication credentials; unauthenticated remote attackers cannot exploit this flaw. Environments using RES for collaborative research, high-performance computing, or shared development infrastructure are most at risk.

Exploitability

Exploitation requires valid RES user credentials—an attacker must have a legitimate account or access to one. No special network position or user interaction is needed beyond having authentication. The attack is straightforward: log in, navigate to the SSH key management interface or API, and replace the private key with a symlink to a target file. Because the operation is performed through standard RES APIs without client-side validation, tools like curl or the RES CLI make this trivial to automate. The CVSS score of 6.5 reflects the medium severity: confidentiality impact is high, but the attack requires prior authentication. This is not an unauthenticated remote code execution; it's a dangerous but bounded privilege-escalation and information-disclosure risk.

Remediation

Upgrade to AWS RES version 2026.06 or later. The vendor recommends this version as the fix. During patching, restrict access to the RES cluster-manager instance to trusted networks where possible. Review and rotate SSH keys and secrets for all users who have had access to affected RES deployments. Audit authentication logs and API access records to detect whether the Auth.GetUserPrivateKey API was called with unusual file paths, which would indicate exploitation attempts.

Patch guidance

Apply the upgrade to RES version 2026.06 as soon as feasible. Verify the patch through your AWS deployment mechanism (CloudFormation, CDK, or manual instance updates, depending on your RES setup). Test the upgrade in a non-production environment first to ensure compatibility with your research workloads. After patching, confirm that the cluster-manager service has restarted and is running the patched version. Check AWS Systems Manager or CloudTrail logs to confirm the deployment completed successfully.

Detection guidance

Monitor Auth.GetUserPrivateKey API calls and authentication logs for unusual file paths being requested—specifically, look for requests that reference paths outside ~/.ssh or that contain symbolic link indicators. Implement file-integrity monitoring on the cluster-manager EC2 instance to detect when ~/.ssh/id_rsa is replaced with a symlink. Review CloudTrail logs for authentication API activity from unexpected source IPs or at unusual times. Check the RES audit logs for private key access or retrieval events that correlate with suspicious user activity. Use osquery or Amazon Inspector to audit symbolic links in user home directories on the cluster-manager instance.

Why prioritize this

Although the CVSS score is medium (6.5), this vulnerability should be prioritized for patching because it directly enables credential theft and lateral movement in environments where RES is trusted with sensitive data. The requirement for authentication is not a strong control in practice—compromised accounts, insider threats, or shared accounts in research environments make this barrier weak. The root-level execution context of the cluster-manager amplifies the blast radius of exploitation. In organizations running RES for intellectually sensitive work or multi-user environments, this is a high-priority patch.

Risk score, explained

The CVSS 3.1 score of 6.5 (MEDIUM) reflects: Attack Vector = Network (AV:N), Attack Complexity = Low (AC:L), Privileges Required = Low (PR:L, authenticated user), User Interaction = None (UI:N), Scope = Unchanged (S:U), Confidentiality = High (C:H), Integrity = None (I:N), Availability = None (A:N). The score appropriately penalizes the requirement for authentication but reflects the severity of unrestricted file-read access as root. In environments where RES user accounts are numerous, widely shared, or access-controlled through federated identity, the practical risk may exceed the baseline score. Conversely, heavily restricted RES deployments with few authenticated users may face lower real-world risk. Apply organizational context when setting patch timelines.

Frequently asked questions