CVE-2026-32055: OpenClaw path traversal enables

CISO Take

OpenClaw versions prior to 2026.2.26 contain a path traversal flaw (CWE-22) in workspace boundary validation that lets a low-privilege attacker write files to arbitrary host filesystem locations by planting in-workspace symlinks targeting non-existent out-of-root paths. With a CVSS of 7.6, low attack complexity, and no user interaction required, this vulnerability ranks in the top 76th EPSS percentile—a realistic exploitation profile despite no current public exploit or KEV listing. AI agent deployments are particularly exposed: OpenClaw agents typically operate with broad filesystem access to perform agentic tasks, so arbitrary file write can translate directly into cron persistence, agent config tampering, or overwriting of model artifacts the agent loads. Patch to 2026.2.26 immediately; if patching is delayed, restrict symlink resolution in agent workspaces using OS-level controls and audit existing workspace directories for anomalous symlinks.

Sources: NVD EPSS GitHub Advisory ATLAS VulnCheck

What is the risk?

High. The CVSS 7.6 score reflects an exploitable combination of network accessibility, low privilege requirement, and no user interaction—meaning any authenticated user of an OpenClaw deployment can trigger the flaw programmatically. AI agent frameworks like OpenClaw commonly run with elevated OS permissions to support agentic task execution, amplifying arbitrary file write impact well beyond what the same vulnerability class yields in a typical web application context. While the absolute EPSS probability is modest (0.32%), ranking in the top 76th percentile indicates statistical risk patterns consistent with exploited vulnerabilities. The AIID #1368 precedent—malicious skills delivering credential stealers through OpenClaw's ecosystem—confirms this attack surface is already under active adversarial attention.

How does the attack unfold?

Initial Access

Attacker obtains low-privilege authenticated access to an OpenClaw deployment via legitimate account creation or by publishing a malicious skill to the OpenClaw skills ecosystem.

AML.T0049

Symlink Staging

Attacker plants an in-workspace symlink pointing to a non-existent out-of-root target path (e.g., /etc/cron.d/backdoor), exploiting the boundary validator's failure to catch escapes via non-existent alias targets.

AML.T0053

Workspace Escape

OpenClaw performs a write operation through the symlinked path; the flawed boundary check resolves the alias without detecting the out-of-root target, causing the write to land on the host filesystem outside the workspace.

AML.T0105

Host Persistence

Attacker's payload—cron job, SSH key injection, agent config override, or corrupted model artifact—is written to an arbitrary host path, establishing persistence or enabling lateral movement from the agent host.

AML.T0081

Initial Access

Attacker obtains low-privilege authenticated access to an OpenClaw deployment via legitimate account creation or by publishing a malicious skill to the OpenClaw skills ecosystem.

AML.T0049

Symlink Staging

Attacker plants an in-workspace symlink pointing to a non-existent out-of-root target path (e.g., /etc/cron.d/backdoor), exploiting the boundary validator's failure to catch escapes via non-existent alias targets.

AML.T0053

Workspace Escape

OpenClaw performs a write operation through the symlinked path; the flawed boundary check resolves the alias without detecting the out-of-root target, causing the write to land on the host filesystem outside the workspace.

AML.T0105

Host Persistence

Attacker's payload—cron job, SSH key injection, agent config override, or corrupted model artifact—is written to an arbitrary host path, establishing persistence or enabling lateral movement from the agent host.

AML.T0081

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

7.6 / 10

EPSS

0.3%

chance of exploitation in 30 days

Higher than 25% of all CVEs

Source: EPSS v3 — FIRST.org

Exploitation Status

No known exploitation

Sophistication

Moderate

What is the attack surface?

AV Network

AC Low

PR Low

UI None

S Unchanged

C Low

I High

A Low

What should I do?

5 steps

Patch: Upgrade OpenClaw to ≥2026.2.26 immediately; both fix commits (46eba86, 1aef45b) address the boundary validation logic.
Workaround if patching is delayed: mount workspace directories with nosymfollow on Linux to prevent symlink escape at the OS layer.
Isolation: Run OpenClaw agents in containers with read-only bind mounts for all directories outside the designated workspace.
Detection: Audit workspace directories for suspicious symlinks—run find /workspace -type l -exec sh -c 'target=$(readlink -f "$1"); echo "$1 -> $target"' _ {} \; and flag any targets outside /workspace.
Monitor: Enable auditd or EDR file creation events for paths outside OpenClaw workspace directories; alert on writes to /etc, /home, /root, or system binary paths.

What does CISA's SSVC say?

Decision Track

Exploitation none

Automatable No

Technical Impact partial

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

How is it classified?

Code Execution Auth Bypass Agent Framework AML.T0049 - Exploit Public-Facing Application AML.T0053 - AI Agent Tool Invocation AML.T0081 - Modify AI Agent Configuration AML.T0105 - Escape to Host

Which compliance frameworks are affected?

This CVE is relevant to:

EU AI Act

Article 15(1) - Accuracy, robustness and cybersecurity

ISO 42001

A.6.2.5 - AI system security

NIST AI RMF

MANAGE-2.2 - Risk Treatment

OWASP LLM Top 10

LLM08 - Excessive Agency

Frequently Asked Questions

What is CVE-2026-32055?

OpenClaw versions prior to 2026.2.26 contain a path traversal flaw (CWE-22) in workspace boundary validation that lets a low-privilege attacker write files to arbitrary host filesystem locations by planting in-workspace symlinks targeting non-existent out-of-root paths. With a CVSS of 7.6, low attack complexity, and no user interaction required, this vulnerability ranks in the top 76th EPSS percentile—a realistic exploitation profile despite no current public exploit or KEV listing. AI agent deployments are particularly exposed: OpenClaw agents typically operate with broad filesystem access to perform agentic tasks, so arbitrary file write can translate directly into cron persistence, agent config tampering, or overwriting of model artifacts the agent loads. Patch to 2026.2.26 immediately; if patching is delayed, restrict symlink resolution in agent workspaces using OS-level controls and audit existing workspace directories for anomalous symlinks.

Is CVE-2026-32055 actively exploited?

No confirmed active exploitation of CVE-2026-32055 has been reported, but organizations should still patch proactively.

How to fix CVE-2026-32055?

1. Patch: Upgrade OpenClaw to ≥2026.2.26 immediately; both fix commits (46eba86, 1aef45b) address the boundary validation logic. 2. Workaround if patching is delayed: mount workspace directories with nosymfollow on Linux to prevent symlink escape at the OS layer. 3. Isolation: Run OpenClaw agents in containers with read-only bind mounts for all directories outside the designated workspace. 4. Detection: Audit workspace directories for suspicious symlinks—run `find /workspace -type l -exec sh -c 'target=$(readlink -f "$1"); echo "$1 -> $target"' _ {} \;` and flag any targets outside /workspace. 5. Monitor: Enable auditd or EDR file creation events for paths outside OpenClaw workspace directories; alert on writes to /etc, /home, /root, or system binary paths.

What systems are affected by CVE-2026-32055?

This vulnerability affects the following AI/ML architecture patterns: agent frameworks, autonomous AI pipelines, CI/CD AI automation, AI tool orchestration.

What is the CVSS score for CVE-2026-32055?

CVE-2026-32055 has a CVSS v3.1 base score of 7.6 (HIGH). The EPSS exploitation probability is 0.32%.

What is the AI security impact?

Affected AI Architectures

agent frameworksautonomous AI pipelinesCI/CD AI automationAI tool orchestration

MITRE ATLAS Techniques

AML.T0049 Exploit Public-Facing Application

AML.T0053 AI Agent Tool Invocation

AML.T0081 Modify AI Agent Configuration

AML.T0105 Escape to Host

Compliance Controls Affected

EU AI Act: Article 15(1)

ISO 42001: A.6.2.5

NIST AI RMF: MANAGE-2.2

OWASP LLM Top 10: LLM08

What are the technical details?

Original Advisory

OpenClaw versions prior to 2026.2.26 contain a path traversal vulnerability in workspace boundary validation that allows attackers to write files outside the workspace through in-workspace symlinks pointing to non-existent out-of-root targets. The vulnerability exists because the boundary check improperly resolves aliases, permitting the first write operation to escape the workspace boundary and create files in arbitrary locations.

Exploitation Scenario

An attacker with low-privilege access to an OpenClaw instance—obtained via a legitimate account or by publishing a malicious skill (analogous to the AIID #1368 attack pattern)—creates a symlink inside the workspace: `ln -s /etc/cron.d/agent_backdoor workspace/tool_cache`. Because /etc/cron.d/agent_backdoor does not yet exist, OpenClaw's boundary validator incorrectly resolves the alias and permits the write. The attacker then triggers any operation that causes OpenClaw to write to workspace/tool_cache (e.g., saving tool output, caching a response). The write lands at /etc/cron.d/agent_backdoor with attacker-controlled content, establishing a cron job that downloads and executes a payload. In fully automated agentic pipelines, this can be triggered without any further attacker interaction after symlink placement, achieving host persistence from a single low-privilege operation.

Weaknesses (CWE)

CWE-22 Improper Limitation of a Pathname to a Restricted Directory ('Path Traversal')

CWE-22 — Improper Limitation of a Pathname to a Restricted Directory ('Path Traversal'): The product uses external input to construct a pathname that is intended to identify a file or directory that is located underneath a restricted parent directory, but the product does not properly neutralize special elements within the pathname that can cause the pathname to resolve to a location that is outside of the restricted directory.

[Implementation] Assume all input is malicious. Use an "accept known good" input validation strategy, i.e., use a list of acceptable inputs that strictly conform to specifications. Reject any input that does not strictly conform to specifications, or transform it into something that does. When performing input validation, consider all potentially relevant properties, including length, type of input, the full range of acceptable values, missing or extra inputs, syntax, consistency across related fields, and conformance to business rules. As an example of business rule logic, "boat" may be syntactically valid because it only contains alphanumeric characters, but it is not valid if the input is only expected to contain colors such as "red" or "blue." Do not rely exclusively on looking for malicious or malformed inputs. This is likely to miss at least one undesirable input, especially if the code's environment changes. This can give attackers enough room to bypass the intended validation. However, denylis
[Architecture and Design] For any security checks that are performed on the client side, ensure that these checks are duplicated on the server side, in order to avoid CWE-602. Attackers can bypass the client-side checks by modifying values after the checks have been performed, or by changing the client to remove the client-side checks entirely. Then, these modified values would be submitted to the server.

Source: MITRE CWE corpus.