CVE-2026-62193: OpenClaw auth bypass

CISO Take

OpenClaw's plugin install wrappers can skip the install policy check, letting a lower-trust caller or an unvetted input path install or persist actions that exceed the authorization the operator intended to grant. The integrity-only impact (confidentiality and availability unaffected) and CVSS 6.5 reflect a policy-enforcement gap gated behind a specific feature being enabled and reachable, not a broad remote takeover. EPSS sits at just 0.2% (top 90th percentile, low near-term exploitation likelihood), there is no CISA KEV listing, no public exploit, and no Nuclei template, so this is not being mass-exploited today. It still deserves priority: OpenClaw carries 447 other CVEs and a package risk score of 0/100, and the AI Incident Database already documents real-world abuse of OpenClaw's plugin/skill ecosystem (AIID #1368, malicious ClawHub skills delivering an infostealer), so authorization gaps in the install path fit a pattern rather than an isolated bug. Patch to 2026.6.9 or later; if immediate patching isn't possible, restrict or disable the plugin install feature for any caller that isn't fully trusted and audit recent plugin installs for anything unauthorized.

Sources: NVD GitHub Advisory EPSS ATLAS VulnCheck

What is the risk?

Medium severity (CVSS 6.5, AV:N/AC:L/PR:L/UI:N) with an integrity-only impact — no confidentiality or availability loss. Exploitability requires the affected install path to be enabled and reachable by a lower-trust caller or attacker-influenced input, so real-world risk is deployment-dependent rather than universal. EPSS (0.2%, ~90th percentile) indicates low predicted exploitation activity, and there is no KEV listing, public exploit, or scanner template. However, the package's poor overall track record (447 other CVEs, 0/100 risk score) and documented prior abuse of its plugin ecosystem (AIID #1368) argue for treating authorization bugs in this subsystem as higher priority than the raw score suggests.

How does the attack unfold?

Entry point

A lower-trust caller or externally-influenced input path reaches OpenClaw's plugin install feature when it is enabled and network-reachable.

AML.T0053

Authorization bypass

The plugin install wrapper fails to enforce the install policy check (CWE-863), letting the request proceed without proper authorization.

AML.T0081

Persistence

The unauthorized plugin is installed and persists in the agent's configuration, retaining capability beyond the caller's intended trust level.

AML.T0110

Impact

The persisted plugin executes or acts within the agent's broader trust context, allowing actions the original caller was never authorized to perform.

Entry point

A lower-trust caller or externally-influenced input path reaches OpenClaw's plugin install feature when it is enabled and network-reachable.

AML.T0053

Authorization bypass

The plugin install wrapper fails to enforce the install policy check (CWE-863), letting the request proceed without proper authorization.

AML.T0081

Persistence

The unauthorized plugin is installed and persists in the agent's configuration, retaining capability beyond the caller's intended trust level.

AML.T0110

Impact

The persisted plugin executes or acts within the agent's broader trust context, allowing actions the original caller was never authorized to perform.

What systems are affected?

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

Do you use OpenClaw? You're affected.

How severe is it?

CVSS 3.1

6.5 / 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 None

I High

A None

What should I do?

1 step

Upgrade OpenClaw to 2026.6.9 or later, where the install policy check is enforced correctly. If immediate patching isn't feasible, disable the plugin/skill install feature entirely or restrict it to callers with full administrative trust; ensure no externally-reachable or lower-privilege path can trigger plugin installation. Audit existing installed plugins for anything not explicitly authorized, and add logging/alerting on plugin install events going forward. Review network exposure of any endpoint that fronts the install wrapper and apply network-level access controls as a compensating measure.

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?

Auth Bypass Supply Chain Agent Plugin AML.T0053 - AI Agent Tool Invocation AML.T0081 - Modify AI Agent Configuration AML.T0110 - AI Agent Tool Poisoning

Which compliance frameworks are affected?

This CVE is relevant to:

EU AI Act

Article 15 - Accuracy, Robustness and Cybersecurity

NIST AI RMF

MANAGE-2.3 - Mechanisms for third-party/plugin risk are in place

OWASP LLM Top 10

LLM03 - Supply Chain Vulnerabilities LLM08 - Excessive Agency

Frequently Asked Questions

What is CVE-2026-62193?

OpenClaw's plugin install wrappers can skip the install policy check, letting a lower-trust caller or an unvetted input path install or persist actions that exceed the authorization the operator intended to grant. The integrity-only impact (confidentiality and availability unaffected) and CVSS 6.5 reflect a policy-enforcement gap gated behind a specific feature being enabled and reachable, not a broad remote takeover. EPSS sits at just 0.2% (top 90th percentile, low near-term exploitation likelihood), there is no CISA KEV listing, no public exploit, and no Nuclei template, so this is not being mass-exploited today. It still deserves priority: OpenClaw carries 447 other CVEs and a package risk score of 0/100, and the AI Incident Database already documents real-world abuse of OpenClaw's plugin/skill ecosystem (AIID #1368, malicious ClawHub skills delivering an infostealer), so authorization gaps in the install path fit a pattern rather than an isolated bug. Patch to 2026.6.9 or later; if immediate patching isn't possible, restrict or disable the plugin install feature for any caller that isn't fully trusted and audit recent plugin installs for anything unauthorized.

Is CVE-2026-62193 actively exploited?

No confirmed active exploitation of CVE-2026-62193 has been reported, but organizations should still patch proactively.

How to fix CVE-2026-62193?

Upgrade OpenClaw to 2026.6.9 or later, where the install policy check is enforced correctly. If immediate patching isn't feasible, disable the plugin/skill install feature entirely or restrict it to callers with full administrative trust; ensure no externally-reachable or lower-privilege path can trigger plugin installation. Audit existing installed plugins for anything not explicitly authorized, and add logging/alerting on plugin install events going forward. Review network exposure of any endpoint that fronts the install wrapper and apply network-level access controls as a compensating measure.

What systems are affected by CVE-2026-62193?

This vulnerability affects the following AI/ML architecture patterns: agent frameworks, plugin/tool ecosystems.

What is the CVSS score for CVE-2026-62193?

CVE-2026-62193 has a CVSS v3.1 base score of 6.5 (MEDIUM). The EPSS exploitation probability is 0.32%.

What is the AI security impact?

Affected AI Architectures

agent frameworksplugin/tool ecosystems

MITRE ATLAS Techniques

AML.T0053 AI Agent Tool Invocation

AML.T0081 Modify AI Agent Configuration

AML.T0110 AI Agent Tool Poisoning

Compliance Controls Affected

EU AI Act: Article 15

NIST AI RMF: MANAGE-2.3

OWASP LLM Top 10: LLM03, LLM08

What are the technical details?

Original Advisory

OpenClaw versions 2026.6.5 before 2026.6.9 contain a vulnerability in the plugin install wrappers that could skip the install policy (authorization) check. When the affected feature is enabled and reachable, a lower-trust caller or a configured input path could execute or persist actions beyond the caller's intended authorization. Impact depends on the operator's configuration and whether lower-trust input can reach the affected path. The issue is fixed in 2026.6.9.

Exploitation Scenario

An organization runs OpenClaw with the plugin install feature enabled to let a semi-trusted internal service or automation pipeline register new skills. Because the install wrapper's authorization check can be skipped, that lower-trust caller — or a configured input path that ingests external data — triggers plugin installation without passing the intended policy gate. The resulting plugin persists in the agent's configuration and executes with the broader trust level of the agent itself, effectively letting the lower-trust caller perform actions it was never authorized for. This mirrors the risk pattern already observed with malicious ClawHub skills (AIID #1368), where the plugin/skill ecosystem was abused to plant unauthorized, persistent functionality.

Weaknesses (CWE)

CWE-863 Incorrect Authorization Primary CWE-863 Incorrect Authorization

CWE-863 — Incorrect Authorization: The product performs an authorization check when an actor attempts to access a resource or perform an action, but it does not correctly perform the check.

[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.