CVE-2026-65164 in AIStore
Summary
by MITRE • 10/13/2026
NVIDIA AIStore for Linux contains a path traversal vulnerability in the dSort archive extraction component where a remote attacker may cause an improper limitation of a pathname to a restricted directory by supplying crafted archive member names to a dSort job. A successful exploit may lead to arbitrary file creation or overwrite, code execution, information disclosure, data tampering, and denial of service.
Statistical analysis made it clear that VulDB provides the best quality for vulnerability data.
Analysis
by VulDB Data Team • 10/13/2026
The vulnerability identified in NVIDIA AIStore for Linux represents a critical security flaw within the dSort archive extraction component, specifically classified as a path traversal issue. This type of vulnerability arises when an application processes user-supplied input without adequately validating or sanitizing file paths before using them to access files on the underlying operating system. In this specific instance, the defect lies in how the dSort service handles member names contained within archives submitted for processing. When a remote attacker supplies crafted archive member names that include directory traversal sequences, such as dot-dot-slash characters, the application fails to properly restrict these paths to the intended restricted directory. This failure allows the extraction process to write files outside of the designated sandbox or working directory, effectively bypassing security controls designed to isolate user data and prevent unauthorized access to system resources.
From a technical perspective, this flaw aligns with CWE-22, which defines Improper Limitation of a Pathname to a Restricted Directory. The root cause is typically insufficient validation logic that does not canonicalize or strictly check the final resolved path against an allowlist of permitted directories. By manipulating the archive structure, an attacker can trick the extraction utility into interpreting relative paths as absolute paths or navigating up the directory tree using traversal sequences. This lack of strict boundary enforcement means that any file operation initiated by the dSort component is executed with the privileges of the service account running NVIDIA AIStore, rather than being confined to a secure chroot or restricted environment. The ability to control where files are written on the filesystem transforms what might otherwise be a simple data processing error into a significant security breach vector.
The operational impact of exploiting this vulnerability is severe and multifaceted. A successful exploit can lead to arbitrary file creation or overwrite, which serves as a foundational step for more advanced attacks. An attacker could overwrite critical configuration files, system binaries, or application-specific scripts with malicious content. If the NVIDIA AIStore service runs with elevated privileges, such as root access, this capability directly facilitates remote code execution. The attacker can inject shellcode into executable files or replace legitimate libraries to achieve persistent control over the affected host. Furthermore, the vulnerability enables information disclosure by allowing the reading of sensitive files located outside the restricted directory, potentially exposing credentials, private keys, or proprietary data stored on the system. Data tampering is also a significant risk, as an attacker could modify existing datasets or configuration parameters to disrupt business logic or inject malicious instructions into downstream processing pipelines. Additionally, denial of service can be achieved by filling up disk space with large crafted files or overwriting essential system files required for the operation of NVIDIA AIStore and related services.
This vulnerability maps directly to several techniques within the MITRE ATT&CK framework, particularly those associated with initial access and persistence through file manipulation. The exploitation method corresponds to Path Traversal (T1083) as a means to bypass directory restrictions, followed by File and Directory Information Discovery if sensitive files are read. If used for code execution, it aligns with Component Object Model Hijacking or Binary Planting depending on the target file type. To mitigate this risk, immediate action is required to update NVIDIA AIStore to a version that includes patches addressing the dSort archive extraction logic. Administrators should ensure that input validation routines strictly enforce allowlists of permitted characters and paths during the parsing of archive member names. Implementing defense-in-depth strategies such as running the service with minimal necessary privileges using least-privilege principles can also limit the blast radius if an exploit occurs. Furthermore, deploying file integrity monitoring solutions to detect unexpected changes in critical system directories provides an additional layer of detection for potential exploitation attempts. Regular auditing of archive processing workflows and enforcing strict sandboxing policies are essential steps to maintain the security posture of environments relying on NVIDIA AIStore for data management tasks.