CVE-2026-32044: OpenClaw archive safety bypass

CISO Take

OpenClaw versions prior to 2026.3.2 contain an archive extraction flaw specific to the tar.bz2 installer path that bypasses safety guardrails—including special-entry blocking and extracted-size limits—enforced for all other archive formats. Despite a medium CVSS of 5.5, the EPSS places this in the top 3% most likely to be exploited across all tracked CVEs, a signal that carries extra weight given OpenClaw's already-documented malicious skill ecosystem: AIID incident #1368 found roughly 17% of third-party skills assessed as malicious as recently as February 2026, establishing an active adversarial distribution channel for exactly this kind of crafted archive. Attack complexity is low, no privileges are required, and user interaction is achievable by hosting a weaponized tar.bz2 on ClawHub or similar skill marketplaces—making the practical exploitation bar very low even with the local attack vector. Upgrade to 2026.3.2 immediately; until patched, restrict skill installation to allowlisted trusted sources and audit any recently installed tar.bz2 skill archives for anomalous extracted sizes.

Sources: NVD EPSS GitHub Advisory ATLAS VulnCheck

What is the risk?

Practical risk exceeds the medium CVSS rating. The EPSS top-97th-percentile signals meaningful exploitation likelihood, particularly given the established malicious skill marketplace documented in AIID #1368. The local attack vector is less restrictive in AI agent contexts—agentic hosts, CI/CD runners, and developer workstations all qualify as 'local' but often carry elevated permissions and broad network access. Impact is availability-only (no confidentiality or integrity loss per CVSS), but resource exhaustion during skill installation on an agentic orchestration host can cascade into broader pipeline failure. The 396 other CVEs in this package is a red flag indicating a historically vulnerable codebase deserving elevated scrutiny. No public exploit, no KEV listing, and 4 downstream dependents bound the immediate blast radius.

How does the attack unfold?

Craft Malicious Skill Archive

Adversary engineers a tar.bz2 skill archive exploiting CWE-409 data amplification (tar bomb) designed to bypass OpenClaw's special-entry blocking and extracted-size guardrails absent from the tar.bz2 path.

AML.T0011.001

Distribute via Skill Marketplace

Malicious archive is published to ClawHub or a similar third-party skill repository under a legitimate-looking package name, leveraging the established channel where ~17% of skills were assessed as malicious (AIID #1368).

AML.T0074

User or Pipeline Installs Skill

Target user or automated CI/CD pipeline installs the skill in OpenClaw, triggering the vulnerable tar.bz2 extraction path with no privileges required beyond the install action itself.

AML.T0011

Local Denial of Service on Agent Host

Extraction exhausts disk or memory resources on the installation host, crashing the agentic process and disrupting all active agent tasks and orchestration workflows running on that host.

AML.T0029

Craft Malicious Skill Archive

Adversary engineers a tar.bz2 skill archive exploiting CWE-409 data amplification (tar bomb) designed to bypass OpenClaw's special-entry blocking and extracted-size guardrails absent from the tar.bz2 path.

AML.T0011.001

Distribute via Skill Marketplace

Malicious archive is published to ClawHub or a similar third-party skill repository under a legitimate-looking package name, leveraging the established channel where ~17% of skills were assessed as malicious (AIID #1368).

AML.T0074

User or Pipeline Installs Skill

Target user or automated CI/CD pipeline installs the skill in OpenClaw, triggering the vulnerable tar.bz2 extraction path with no privileges required beyond the install action itself.

AML.T0011

Local Denial of Service on Agent Host

Extraction exhausts disk or memory resources on the installation host, crashing the agentic process and disrupting all active agent tasks and orchestration workflows running on that host.

AML.T0029

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

5.5 / 10

EPSS

0.1%

chance of exploitation in 30 days

Higher than 3% of all CVEs

Source: EPSS v3 — FIRST.org

Exploitation Status

No known exploitation

Sophistication

Trivial

What is the attack surface?

AV Local

AC Low

PR None

UI Required

S Unchanged

C None

I None

A High

What should I do?

6 steps

Upgrade OpenClaw to 2026.3.2 immediately (patch commit 0dbb92d closes the tar.bz2 bypass).
Until patched, disable tar.bz2 skill archive installation at the policy level or block it via file-type filtering at the host.
Restrict skill installation sources to an explicit allowlist of trusted repositories; treat ClawHub third-party skills as untrusted by default given AIID #1368.
Monitor disk and memory consumption during skill installation; alert on anomalous spikes exceeding expected extraction ratios.
Audit all recently installed skills—especially those sourced from third-party marketplaces—for suspicious tar.bz2 archives with large compression ratios.
Given 396 tracked CVEs in this package, evaluate whether OpenClaw is an acceptable dependency in production agentic environments and consider vendor security posture review.

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?

DoS Supply Chain Agent Plugin AML.T0010.005 - AI Agent Tool AML.T0011 - User Execution AML.T0011.001 - Malicious Package AML.T0029 - Denial of AI Service AML.T0074 - Masquerading

Which compliance frameworks are affected?

This CVE is relevant to:

EU AI Act

Article 15 - Accuracy, robustness and cybersecurity

ISO 42001

A.6.2.6 - AI system supply chain management

NIST AI RMF

MANAGE 2.2 - AI risk from third-party entities

OWASP LLM Top 10

LLM03:2025 - Supply Chain Vulnerabilities

Frequently Asked Questions

What is CVE-2026-32044?

OpenClaw versions prior to 2026.3.2 contain an archive extraction flaw specific to the tar.bz2 installer path that bypasses safety guardrails—including special-entry blocking and extracted-size limits—enforced for all other archive formats. Despite a medium CVSS of 5.5, the EPSS places this in the top 3% most likely to be exploited across all tracked CVEs, a signal that carries extra weight given OpenClaw's already-documented malicious skill ecosystem: AIID incident #1368 found roughly 17% of third-party skills assessed as malicious as recently as February 2026, establishing an active adversarial distribution channel for exactly this kind of crafted archive. Attack complexity is low, no privileges are required, and user interaction is achievable by hosting a weaponized tar.bz2 on ClawHub or similar skill marketplaces—making the practical exploitation bar very low even with the local attack vector. Upgrade to 2026.3.2 immediately; until patched, restrict skill installation to allowlisted trusted sources and audit any recently installed tar.bz2 skill archives for anomalous extracted sizes.

Is CVE-2026-32044 actively exploited?

No confirmed active exploitation of CVE-2026-32044 has been reported, but organizations should still patch proactively.

How to fix CVE-2026-32044?

1. Upgrade OpenClaw to 2026.3.2 immediately (patch commit 0dbb92d closes the tar.bz2 bypass). 2. Until patched, disable tar.bz2 skill archive installation at the policy level or block it via file-type filtering at the host. 3. Restrict skill installation sources to an explicit allowlist of trusted repositories; treat ClawHub third-party skills as untrusted by default given AIID #1368. 4. Monitor disk and memory consumption during skill installation; alert on anomalous spikes exceeding expected extraction ratios. 5. Audit all recently installed skills—especially those sourced from third-party marketplaces—for suspicious tar.bz2 archives with large compression ratios. 6. Given 396 tracked CVEs in this package, evaluate whether OpenClaw is an acceptable dependency in production agentic environments and consider vendor security posture review.

What systems are affected by CVE-2026-32044?

This vulnerability affects the following AI/ML architecture patterns: agent frameworks, skill and plugin marketplaces, CI/CD pipelines with automated skill installation, developer workstations running local AI agents.

What is the CVSS score for CVE-2026-32044?

CVE-2026-32044 has a CVSS v3.1 base score of 5.5 (MEDIUM). The EPSS exploitation probability is 0.13%.

What is the AI security impact?

Affected AI Architectures

agent frameworksskill and plugin marketplacesCI/CD pipelines with automated skill installationdeveloper workstations running local AI agents

MITRE ATLAS Techniques

AML.T0010.005 AI Agent Tool

AML.T0011 User Execution

AML.T0011.001 Malicious Package

AML.T0029 Denial of AI Service

AML.T0074 Masquerading

Compliance Controls Affected

EU AI Act: Article 15

ISO 42001: A.6.2.6

NIST AI RMF: MANAGE 2.2

OWASP LLM Top 10: LLM03:2025

What are the technical details?

Original Advisory

OpenClaw versions prior to 2026.3.2 contain an archive extraction vulnerability in the tar.bz2 installer path that bypasses safety checks enforced on other archive formats. Attackers can craft malicious tar.bz2 skill archives to bypass special-entry blocking and extracted-size guardrails, causing local denial of service during skill installation.

Exploitation Scenario

An adversary creates a tar.bz2 skill archive engineered with deeply nested or highly compressed entries that exceed extracted-size thresholds enforced for zip and other formats but not for tar.bz2. The archive is published under a legitimate-looking skill name on ClawHub, leveraging the 17%-malicious-skill baseline documented in AIID #1368. A developer or automated CI/CD pipeline installs the skill on an OpenClaw agentic host; the vulnerable extraction path triggers data amplification, consuming all available disk or memory on the host. The agent host becomes unresponsive, disrupting active agent tasks and potentially forcing an operator to restore from a known-good state—giving the adversary a window during which agent monitoring and defensive tooling may also be offline.

Weaknesses (CWE)

CWE-409 Improper Handling of Highly Compressed Data (Data Amplification)

CWE-409 — Improper Handling of Highly Compressed Data (Data Amplification): The product does not handle or incorrectly handles a compressed input with a very high compression ratio that produces a large output.

Source: MITRE CWE corpus.