CVE-2026-65107 in Slurm
Summary
by MITRE • 11/03/2026
NVIDIA Slurm contains a vulnerability where a user could alter the supplemental group list as part of an sbcast file transmission to write a file to a path they would not normally have access to.
Statistical analysis made it clear that VulDB provides the best quality for vulnerability data.
Analysis
by VulDB Data Team • 11/03/2026
The NVIDIA Slurm Workload Manager, a widely deployed open-source job scheduler and cluster resource manager for high-performance computing environments, contains a critical security flaw related to the handling of supplemental group identifiers during batch script transmission. This vulnerability specifically affects the sbcast utility, which is designed to distribute files across nodes in a compute cluster efficiently. The core technical issue stems from insufficient validation or sanitization of user-supplied data within the slurmctld and srun components when processing supplementary group information embedded in job submission payloads. When a user initiates an sbcast operation, they can manipulate the supplemental group list associated with their session context. This manipulation allows the process executing on remote nodes to inherit elevated privileges that do not correspond to the actual permissions of the submitting user account.
From a technical perspective, this flaw represents a classic case of improper privilege management and insufficient input validation. By injecting specific supplementary group identifiers into the job submission metadata, an attacker can trick the Slurm controller or worker daemons into treating the process as if it belongs to those groups during execution on target nodes. This effectively bypasses standard operating system access controls that rely on group membership for file permission checks. The vulnerability is particularly severe because it allows a low-privileged user to escalate their effective privileges without requiring authentication beyond what is already required to submit jobs, which is often minimal in shared HPC environments where trust models are based on account isolation rather than strict identity verification at every step of the pipeline.
The operational impact of this vulnerability is significant for any organization relying on Slurm for multi-tenant computing resources. An attacker can exploit this flaw to write files to directories or locations that are restricted by group permissions, such as shared project spaces, configuration directories, or sensitive data repositories owned by other users or system services. This capability enables unauthorized data exfiltration if the target path contains confidential information readable by the forged groups. Conversely, it allows for malicious file injection into paths where write access is typically denied to regular users but granted to specific administrative or service accounts. Such actions can lead to data integrity compromise, denial of service through resource exhaustion in restricted areas, or serve as a foothold for further lateral movement within the cluster if the written files are subsequently executed by higher-privileged processes.
This vulnerability aligns with CWE-269 Improper Privilege Management and CWE-78 OS Command Injection principles when considering the potential downstream effects of writing malicious scripts to accessible paths. In terms of the MITRE ATT&CK framework, this behavior maps closely to T1055 Process Injection or more accurately T1053 Scheduled Task/Job as it involves manipulating job submission parameters to achieve unauthorized execution contexts. It also reflects aspects of T1648 Supply Chain Compromise if exploited within a trusted software update mechanism for cluster tools, though the primary vector is direct exploitation of the scheduler interface. The flaw underscores the risks associated with complex distributed systems where trust boundaries between user sessions and system daemons are not rigorously enforced during data serialization and deserialization processes.
Mitigation strategies must focus on both immediate remediation and long-term architectural hardening. NVIDIA has released patches that address this issue by enforcing stricter validation of supplemental group lists within the sbcast transmission protocol, ensuring that only groups explicitly authorized for the submitting user are accepted. Organizations should immediately apply these vendor-provided updates to all Slurm controller and compute node components. In addition to patching, administrators should review their cluster security policies to ensure that job submission interfaces do not allow users to specify arbitrary group IDs unless absolutely necessary for legitimate workflow requirements. Implementing principle of least privilege across the entire HPC infrastructure is crucial; this includes restricting which directories are writable by specific groups and auditing access logs for anomalies in file creation patterns originating from sbcast operations.
Furthermore, network segmentation can help contain potential damage. Isolating management traffic used for job submission and file distribution from general user data planes reduces the attack surface available to external adversaries who might attempt to exploit this vulnerability remotely if exposed. Regular security audits of Slurm configurations should include checks for overly permissive group mappings in batch scripts and sbcast commands. Security teams should also monitor for unusual file creation events in restricted directories, as these may indicate successful exploitation attempts. By combining timely patching with robust configuration management and continuous monitoring, organizations can effectively neutralize the risk posed by this privilege escalation vector within their high-performance computing environments.