/

CVE-2026-100587

Severity:

HIGH RISK

(8.8)

Missing Authorization (CWE-862) in OpenClaw Codex Command

CVE-2026-100587

Deferred

Vulnerability type

Category:

Missing Authorization

CWE:

CWE-862

Impact:

Unauthorized plugin installation and privileged process execution

Severity:

HIGH RISK

(8.8)

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

OpenClaw does not properly verify owner authorization for the Codex computer-use installation command. A non-owner channel sender may install arbitrary plugins and execute MCP processes with the privileges of the OpenClaw user, creating risks to host confidentiality, integrity, and availability.

Affected context

The issue occurs in OpenClaw deployments that expose the Codex computer-use installation command through channels where multiple senders may interact. It becomes risky when authorization is not independently enforced for each command request and a non-owner sender can reach the installation workflow.

Exploit preconditions

  • Attack surface: OpenClaw channel command handling and the Codex computer-use installation workflow - Required access: Ability to send messages or commands through an enabled OpenClaw channel - Trigger point: Processing of the plugin installation command - Why exploitation is possible: The command path does not reliably confirm that the sender has owner-level authorization before permitting plugin installation or MCP process execution

Why this happens

The underlying bug is an authorization failure in which a sensitive administrative operation relies on insufficient or missing checks for the identity and privileges of the requesting channel sender. This maps to CWE-862, Missing Authorization, because the application permits a protected action without consistently enforcing the required access-control decision.

Potential impact

  • An unauthorized sender may install arbitrary plugins.
  • An attacker could execute MCP processes with OpenClaw user privileges.
  • Host data confidentiality could be reduced.
  • Host files or application state could be modified.
  • Service availability could be disrupted.
  • Installed malicious components could support persistence or further compromise.

General mitigation guidance

Mitigation generally focuses on enforcing authorization at the command execution boundary rather than relying on channel context or user-interface restrictions. Sensitive installation and process-launch operations should use explicit owner or administrator checks, deny access by default, and validate the authenticated sender independently for every request. Plugin sources and MCP process definitions can be constrained through allowlists, controlled repositories, and schema validation. Running OpenClaw with least-privilege accounts, restricted filesystem access, and limited process permissions can reduce impact when authorization controls fail.

How teams detect issues like this

Detection can include static analysis and code review for administrative command handlers that lack centralized authorization checks, trust sender metadata without verification, or invoke plugin and process-management functions directly. Runtime monitoring can alert on plugin installations, MCP process launches, and command requests from non-owner identities. Policy enforcement may require approval workflows, signed or allowlisted plugins, and audit records linking each sensitive action to an authenticated principal. Automated security review tools can help identify missing authorization boundaries.

Sources

SourceLink
github.comgithub.com/openclaw/openclaw/security/advisories/GHSA-pjjr-5qhr-5w6r
vulncheck.comvulncheck.com/advisories/openclaw-before-2026.7.1-authorization-bypass-via-codex-install

Frequently asked

questions

A sender who can reach the affected OpenClaw channel command may be able to exploit it if the deployment does not correctly distinguish owner requests from non-owner requests.

No direct host login is necessarily required. The relevant prerequisite is access to a channel that exposes the vulnerable installation command.

Enforce a centralized, deny-by-default authorization check at the command execution boundary and verify the sender's identity and owner privileges for every request.

Apply consistent access-control middleware to administrative commands, restrict plugin and process sources, run services with least privilege, and audit sensitive actions for unexpected senders.

Find out what's already
exploitable in your codebase.