CVE-2026-100596: OpenClaw: missing authz in /mcp set enables persistent RCE

HIGH

Published September 26, 2026

CISO Take

OpenClaw before 2026.7.1 doesn't check ownership when a user runs the /mcp set and /mcp unset commands. Any low-privileged, non-owner user can therefore write an arbitrary stdio MCP command into the agent's configuration, and it runs with the OpenClaw process privileges the next time the configuration loads. This is a network-reachable, low-complexity flaw with no user interaction (CVSS 8.8), and it turns any ordinary account into persistent code execution on the host. Exploitation likelihood is currently modest: EPSS is 0.00246 (top 85th percentile), it is not in CISA KEV, and there is no public exploit or Nuclei template. The exploit steps are trivial once someone has an account, though, so the risk grows quickly in multi-user or shared-channel deployments. Upgrade to OpenClaw 2026.7.1 or later, restrict who can reach the agent's command interface, and audit existing MCP configuration for stdio commands nobody approved.

Sources:NVDEPSSGitHub Advisoryvulncheck.comATLAS

What is the risk?

High. The flaw is a missing authorization check (CWE-862) reachable over the network with low privileges, low complexity and no user interaction. Impact is full compromise of host confidentiality, integrity and availability, because injected stdio MCP commands run as the OpenClaw process. Exploitation likelihood today is moderate to low (EPSS 0.00246, not in KEV, no public PoC), but the attack needs no AI/ML expertise and no special tooling. Exposure is highest where OpenClaw serves multiple users, shared chat channels or teams, and where the agent process holds credentials, API keys or broad filesystem access.

How does the attack unfold?

Initial Access

Attacker authenticates as a low-privileged non-owner user of the OpenClaw instance.

AML.T0012

Exploitation

Attacker sends /mcp set to register a malicious stdio MCP command, bypassing the missing owner authorization check.

AML.T0081

Persistence

The malicious MCP entry is stored in the configuration and survives restarts.

AML.T0110

Impact

When the configuration loads, the command executes with OpenClaw process privileges, compromising the host's confidentiality, integrity and availability.

AML.T0050

Initial Access

Attacker authenticates as a low-privileged non-owner user of the OpenClaw instance.

AML.T0012

Exploitation

Attacker sends /mcp set to register a malicious stdio MCP command, bypassing the missing owner authorization check.

AML.T0081

Persistence

The malicious MCP entry is stored in the configuration and survives restarts.

AML.T0110

Impact

When the configuration loads, the command executes with OpenClaw process privileges, compromising the host's confidentiality, integrity and availability.

AML.T0050

What systems are affected?

Package	Ecosystem	Vulnerable Range	Patched
OpenClaw	pip	—	No patch
3 dependents32% patched~7d to patchFull package profile →

Do you use OpenClaw? You're affected.

How severe is it?

CVSS 3.1

8.8 / 10

EPSS

0.2%

chance of exploitation in 30 days

Higher than 16% of all CVEs

Source: EPSS v3 — FIRST.org

Exploitation Status

No known exploitation

Sophistication

Trivial

What is the attack surface?

AVNetwork

ACLow

PRLow

UINone

SUnchanged

CHigh

IHigh

AHigh

What should I do?

1 step

1) Upgrade OpenClaw to 2026.7.1 or later (see GHSA-wwx7-573h-pqwc and the VulnCheck advisory). 2) Until patched, restrict access to the agent's chat or command interface to trusted owners and disable or block the /mcp set and /mcp unset commands for non-owners where possible. 3) Review current MCP configuration for unknown stdio entries (unexpected command, args or env values) and remove them; treat any unrecognised entry as a potential compromise. 4) Run OpenClaw as an unprivileged user in a container or sandbox with minimal filesystem, network and secret access. 5) Rotate API keys and tokens accessible to the OpenClaw process if tampering is suspected. 6) Detection: alert on config file changes, on /mcp set and /mcp unset invocations by non-owner users, and on child processes spawned by the OpenClaw process that are not on the expected MCP server allowlist.

What does CISA's SSVC say?

DecisionTrack

Exploitationnone

AutomatableNo

Technical Impacttotal

Source: CISA Vulnrichment (SSVC v2.0). Decision based on the CISA Coordinator decision tree.

How is it classified?

Auth BypassCode ExecutionAgentPluginFrameworkAML.T0012- Valid Accounts AML.T0050- Command and Scripting Interpreter AML.T0081- Modify AI Agent Configuration AML.T0110- AI Agent Tool Poisoning AML.T0112.000- Local AI Agent

Which compliance frameworks are affected?

This CVE is relevant to:

EU AI Act

Art. 15-Accuracy, robustness and cybersecurity

ISO 42001

A.6.2.6-AI system operation and monitoring

NIST AI RMF

MEASURE 2.7-AI system security and resilience are evaluated and documented

OWASP LLM Top 10

LLM03-Supply ChainLLM06-Excessive Agency

Frequently Asked Questions

What is CVE-2026-100596?

Is CVE-2026-100596 actively exploited?

No confirmed active exploitation of CVE-2026-100596 has been reported, but organizations should still patch proactively.

How to fix CVE-2026-100596?

1) Upgrade OpenClaw to 2026.7.1 or later (see GHSA-wwx7-573h-pqwc and the VulnCheck advisory). 2) Until patched, restrict access to the agent's chat or command interface to trusted owners and disable or block the /mcp set and /mcp unset commands for non-owners where possible. 3) Review current MCP configuration for unknown stdio entries (unexpected command, args or env values) and remove them; treat any unrecognised entry as a potential compromise. 4) Run OpenClaw as an unprivileged user in a container or sandbox with minimal filesystem, network and secret access. 5) Rotate API keys and tokens accessible to the OpenClaw process if tampering is suspected. 6) Detection: alert on config file changes, on /mcp set and /mcp unset invocations by non-owner users, and on child processes spawned by the OpenClaw process that are not on the expected MCP server allowlist.

What systems are affected by CVE-2026-100596?

This vulnerability affects the following AI/ML architecture patterns: Agent frameworks, MCP tool integrations, Multi-user shared AI assistants, Self-hosted agent gateways.

What is the CVSS score for CVE-2026-100596?

CVE-2026-100596 has a CVSS v3.1 base score of 8.8 (HIGH). The EPSS exploitation probability is 0.25%.

What is the AI security impact?

Affected AI Architectures

Agent frameworksMCP tool integrationsMulti-user shared AI assistantsSelf-hosted agent gateways

MITRE ATLAS Techniques

AML.T0012Valid Accounts

AML.T0050Command and Scripting Interpreter

AML.T0081Modify AI Agent Configuration

AML.T0110AI Agent Tool Poisoning

AML.T0112.000Local AI Agent

Compliance Controls Affected

EU AI Act: Art. 15

ISO 42001: A.6.2.6

NIST AI RMF: MEASURE 2.7

OWASP LLM Top 10: LLM03, LLM06

What are the technical details?

Original Advisory

OpenClaw versions before 2026.7.1 fail to properly authorize non-owner users executing MCP configuration changes through /mcp set and /mcp unset commands. Attackers can persist arbitrary stdio MCP commands that execute with OpenClaw process privileges when configuration loads, compromising host confidentiality, integrity, and availability.

Exploitation Scenario

An organization runs a shared OpenClaw agent in a team chat workspace, with the owner configured as the administrator. A low-privileged member, or an attacker who has compromised one such account, sends '/mcp set' with a new stdio MCP server whose command downloads and runs a reverse shell. The missing owner check lets the change through and it is written to the persistent configuration. When OpenClaw reloads its configuration (restart, redeploy or scheduled reload), the command executes with the agent's privileges. The attacker now holds a foothold on the host, can read LLM API keys and connected-tool credentials, and keeps access across restarts through the planted configuration entry.

Weaknesses (CWE)

CWE-862Missing AuthorizationPrimary CWE-862Missing Authorization

CWE-862 — Missing Authorization: The product does not perform an authorization check when an actor attempts to access a resource or perform an action.

[Architecture and Design] Divide the product into anonymous, normal, privileged, and administrative areas. Reduce the attack surface by carefully mapping roles with data and functionality. Use role-based access control (RBAC) [REF-229] to enforce the roles at the appropriate boundaries. Note that this approach may not protect against horizontal authorization, i.e., it will not protect a user from attacking others with the same role.
[Architecture and Design] Ensure that access control checks are performed related to the business logic. These checks may be different than the access control checks that are applied to more generic resources such as files, connections, processes, memory, and database records. For example, a database may restrict access for medical records to a specific database user, but each record might only be intended to be accessible to the patient and the patient's doctor [REF-7].

Source: MITRE CWE corpus.