/

CVE-2026-90619

Severity:

HIGH RISK

(7.3)

OS Command Injection (CWE-78) in HexStrike Execute

CVE-2026-90619

Deferred

Vulnerability type

Category:

OS Command Injection

CWE:

CWE-77, CWE-78

Impact:

Remote Code Execution

Severity:

HIGH RISK

(7.3)

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

A remotely reachable Execute Endpoint in 0x4m4 HexStrike AI may pass attacker-controlled code or script input to operating-system command execution. An attacker who can reach the endpoint and influence the relevant argument could execute commands with the service's privileges.

Affected context

The vulnerability is associated with the Execute Endpoint implemented in hexstrike_server.py. It becomes risky when the endpoint is network-accessible and accepts code or script content that reaches an operating-system command execution path without sufficient separation between data and commands. The exact deployment exposure depends on network controls, authentication, and the service account's permissions.

Exploit preconditions

  • Attack surface: The remotely accessible Execute Endpoint. - Required access: Network access to the endpoint, plus any authentication or authorization

accepted by the deployment.

  • Trigger point: Processing the endpoint's code or script argument. - Why exploitation is possible: Input may influence an operating-system command execution

operation without adequate command and argument separation.

Why this happens

The underlying bug is an OS command injection weakness. Application-controlled command execution appears to incorporate externally influenced script or code input into a shell or operating-system command path. This maps primarily to CWE-78, Improper Neutralization of Special Elements used in an OS Command. CWE-77 is also relevant because the broader issue involves improper neutralization of command syntax before interpretation by a command processor.

Potential impact

  • An attacker may execute operating-system commands with the service account's privileges.
  • The affected host or container could be compromised.
  • Sensitive files, environment variables, or service credentials could be exposed.
  • An attacker could alter application data or disrupt service availability.
  • Compromise could enable lateral movement where network access and permissions allow it.

General mitigation guidance

Mitigation generally focuses on removing shell interpretation from application-controlled execution paths. Secure-by-default designs use fixed command mappings, structured process APIs, and explicit argument arrays rather than concatenated command strings. Code or script input should be treated as untrusted data, constrained by a strict allowlist or replaced with a purpose-specific execution interface. Where execution is required, isolation, least-privilege service accounts, restricted filesystem access, and limited network permissions can reduce impact. Authentication, authorization, and network policy should limit access to the endpoint, while logging and review should cover all execution requests.

How teams detect issues like this

Detection can include static-analysis rules for shell invocation APIs, string concatenation or interpolation used to build commands, and externally controlled values passed into process execution functions. Code review should examine endpoint arguments, command construction, shell-enabled process options, and validation performed before execution. Dynamic testing can use benign metacharacter and separator probes in isolated environments to verify that inputs remain data rather than command syntax. Policy enforcement may restrict shell execution from network-facing services and require structured process-spawning interfaces; automated security review tools can help identify these patterns.

Sources

SourceLink
github.comgithub.com/0x4m4/hexstrike-ai/issues/222
vuldb.comvuldb.com/cve/CVE-2026-90619
vuldb.comvuldb.com/submit/914810
vuldb.comvuldb.com/vuln/403200
vuldb.comvuldb.com/vuln/403200/cti

Frequently asked

questions

Potentially. The reported attack surface is a remotely reachable Execute Endpoint, but

practical exploitability depends on network exposure, authentication, authorization, and

deployment configuration.

An attacker generally needs network access to the endpoint and whatever permissions are

required to invoke it. The necessary access may be unauthenticated or authenticated

depending on how the service is deployed.

Avoid constructing shell commands from request data. Prefer fixed operations and structured

process APIs with explicit argument lists, combined with strict allowlisting and

least-privilege execution.

Treat all endpoint parameters as untrusted, prohibit unsafe shell invocation patterns,

enforce authorization and network restrictions, and include static analysis, code review,

and isolated dynamic testing in the development process.

Find out what's already
exploitable in your codebase.