/

CVE-2026-65099

Severity:

HIGH RISK

(7.8)

OS Command Injection (CWE-78) in NVIDIA NemoClaw CLI

CVE-2026-65099

Analyzed

Vulnerability type

Category:

OS Command Injection

CWE:

CWE-78

Impact:

Code Execution

Severity:

HIGH RISK

(7.8)

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 NemoClaw for Linux contains an OS command injection vulnerability in its command-line interface. If an attacker can influence CLI input that reaches an unsafe command execution path, successful exploitation may enable code execution, data tampering, information disclosure, or denial of service.

Affected context

The vulnerability occurs in the NVIDIA NemoClaw command-line interface for Linux when externally influenced values are incorporated into operating-system commands without sufficient separation between data and command structure. Risk depends on whether an attacker can supply or modify the relevant CLI input and on the permissions available to the NemoClaw process.

Exploit preconditions

  • Attack surface: NVIDIA NemoClaw CLI command-processing functionality on Linux
  • Required access: Ability to invoke the CLI or influence input consumed by a CLI operation
  • Trigger point: Processing of attacker-influenced input that is passed to an operating-system command
  • Why exploitation is possible: Input may be interpreted as command syntax rather than as data, allowing unintended command execution

Why this happens

The underlying bug is an OS command injection pattern in which untrusted or insufficiently constrained input is combined with a shell command or command interpreter. When command structure and user-controlled data are not separated, special characters or constructed arguments may alter the intended operation. This behavior maps to CWE-78: Improper Neutralization of Special Elements used in an OS Command, commonly known as OS command injection.

Potential impact

  • An attacker may execute commands with the privileges of the NemoClaw process.
  • Sensitive information could be disclosed.
  • Local data or application state could be modified or tampered with.
  • The affected system may experience denial of service.
  • Compromise could enable further actions against accessible resources.

General mitigation guidance

Mitigation generally focuses on removing shell interpretation from command execution paths and treating all external values as untrusted data.

Common defensive patterns include:

  • Prefer direct process-execution APIs that pass an argument array rather than constructing shell command strings.
  • Apply strict allowlists and type-aware validation to values accepted by the CLI.
  • Avoid shell invocation when an equivalent library or native API is available.
  • Separate command selection from user-controlled arguments and enforce fixed command mappings.
  • Run the CLI with the minimum filesystem, network, and operating-system privileges required.
  • Use isolated execution environments and restrictive policies to limit the consequences of command execution.

How teams detect issues like this

Detection approaches may include:

  • Static analysis for shell invocation APIs, command-string concatenation, and subprocess calls receiving unsanitized input.
  • Code review checks for shell metacharacter handling, dynamic command construction, and missing allowlists.
  • Runtime monitoring of unusual child-process creation, unexpected interpreters, and command execution by NemoClaw.
  • Policy enforcement requiring direct argument-based process execution and security review for new command-line operations.
  • Automated security and code-review tools can help identify command-injection patterns before deployment.

Sources

SourceLink
github.comgithub.com/NVIDIA/product-security/tree/main/2026/5872
nvd.nist.govnvd.nist.gov/vuln/detail/CVE-2026-65099
cve.orgcve.org/CVERecord?id=CVE-2026-65099

Frequently asked

questions

Exploitability depends on how NemoClaw is exposed and whether an attacker can invoke the CLI or influence its inputs. The description alone does not establish direct remote exploitation.

An attacker generally needs a way to supply input processed by the vulnerable CLI path, such as local execution access, a reachable wrapper, or another component that forwards attacker-controlled values.

Avoid shell-based command construction, use direct process APIs with separate arguments, and enforce strict allowlists for accepted values.

Treat all CLI and integration inputs as untrusted, apply static analysis and code review rules for dynamic command execution, and run the component with least privilege.

Find out what's already
exploitable in your codebase.