/

CVE-2026-65092

Severity:

HIGH RISK

(8.5)

Path Traversal (CWE-22) in NVIDIA OpenShell Sandbox

CVE-2026-65092

Analyzed

Vulnerability type

Category:

Path Traversal

CWE:

CWE-22

Impact:

Information Disclosure and Data Tampering

Severity:

HIGH RISK

(8.5)

CVSS version:

3.1

Catch this CVE while code is being written

Detect vulnerable code directly in pull requests, so CVEs are fixed before merge.

Summary

NVIDIA OpenShell Sandbox for Linux contains a path traversal vulnerability that can bypass Layer 7 REST network policy controls. An attacker who can submit or influence relevant REST requests may access unintended paths, potentially exposing information or modifying data.

Affected context

The vulnerability occurs in NVIDIA OpenShell Sandbox deployments that enforce Layer 7 REST network policies while handling attacker-influenced resource paths. Risk depends on whether requests can reach the affected policy enforcement path and whether the sandbox process has access to sensitive files or writable data outside the intended REST resource scope.

Exploit preconditions

  • Attack surface: Layer 7 REST request handling in the Linux sandbox
  • Required access: Ability to send or influence requests processed by the affected REST policy
  • Trigger point: Path resolution and authorization of a REST resource
  • Why exploitation is possible: Improper path validation may allow traversal sequences or equivalent path representations to escape the policy's intended resource boundary

Why this happens

The underlying bug is insufficient canonicalization and validation of user-influenced paths before applying access-control decisions. When a system checks a non-canonical path but later resolves it differently, an attacker may reference resources outside the permitted directory or endpoint scope. This behavior maps to CWE-22: Improper Limitation of a Pathname to a Restricted Directory, commonly known as path traversal.

Potential impact

  • An attacker may disclose files or REST-accessible information outside the intended policy scope.
  • An attacker could tamper with writable data reachable through the sandbox.
  • Sensitive configuration, metadata, or application content may be exposed.
  • The affected service may experience integrity loss if unauthorized resources can be modified.

General mitigation guidance

Mitigation generally focuses on enforcing authorization after secure path normalization and limiting requests to an explicit resource scope.

Common defensive patterns include:

  • Canonicalizing paths before authorization decisions.
  • Rejecting traversal sequences, ambiguous encodings, and alternate path representations.
  • Resolving requested resources against a fixed permitted root and verifying the resulting path remains within that root.
  • Preferencing allowlisted REST routes and resource identifiers over filesystem-like path construction.
  • Running the sandbox with least-privilege filesystem permissions and minimizing writable locations.
  • Applying read-only mounts and separation of sensitive data where practical.
  • Treating client-supplied paths as untrusted even when they pass initial format validation.

How teams detect issues like this

Detection approaches commonly include:

  • Static analysis checks for path joins, normalization, and authorization performed in an unsafe order.
  • Code review of REST handlers that concatenate user input into filesystem or resource paths.
  • Tests covering encoded separators, dot segments, nested traversal, symlink resolution, and platform-specific path forms.
  • Runtime logging and policy alerts for rejected traversal patterns and requests resolving outside approved resource roots.
  • Automated security testing and code review tools that identify path traversal and authorization-boundary weaknesses.

Sources

SourceLink
github.comgithub.com/NVIDIA/product-security/tree/main/2026/5872
nvd.nist.govnvd.nist.gov/vuln/detail/CVE-2026-65092
cve.orgcve.org/CVERecord?id=CVE-2026-65092

Frequently asked

questions

It may be exploitable through the affected Layer 7 REST interface when that interface is reachable by an attacker and processes attacker-controlled paths. The practical scope depends on deployment exposure and policy configuration.

An attacker generally needs the ability to send requests to, or otherwise influence requests handled by, the affected sandbox REST path. Authentication and network restrictions may reduce exposure but do not replace correct path validation.

Normalize and resolve paths before authorization, enforce an explicit permitted resource boundary, and reject ambiguous or traversal-based representations. Least-privilege filesystem access further limits consequences.

Use allowlisted routes and resource identifiers, centralize path validation, test canonicalization edge cases, review authorization boundaries, and monitor rejected or anomalous path requests.

Find out what's already
exploitable in your codebase.