CVE-2026-62225: OpenClaw authz bypass

CISO Take

OpenClaw, an AI agent framework, has an authorization bypass in its skill command dispatch logic that lets a lower-trust caller — a sub-agent, delegated skill, or externally-configured plugin operating under a restricted policy — execute or persist actions it was never granted permission for, effectively defeating the tool-policy layer meant to contain it. This matters because agent frameworks deploy tiered trust models specifically so a less-trusted skill can't escalate into full agent privileges; this flaw removes that containment whenever the affected dispatch feature is enabled and network-reachable, and the fix maps to CWE-863 (Incorrect Authorization) rather than a missing check entirely, meaning the policy exists but can be routed around via crafted input paths. There is no EPSS score, no CISA KEV listing, and no public exploit or Nuclei scanner template for this CVE, and the CVSS 5.4 (network vector, low complexity, low privileges required, no user interaction, confidentiality/integrity impact both low, no availability impact) reflects a real but bounded risk rather than an active-exploitation emergency. Upgrade OpenClaw to 2026.5.18 or later now; in the interim, audit which skills/sub-agents run at reduced trust tiers, confirm none can reach the affected dispatch path, and review agent action/audit logs for tool invocations that don't match the caller's assigned privilege level.

Sources: NVD GitHub Advisory ATLAS vulncheck.com

What is the risk?

Medium severity (CVSS 5.4) with low attack complexity and no user interaction, but privileges required (an attacker needs some foothold as a lower-trust caller within the agent system, not unauthenticated network access). No evidence of active exploitation (not in CISA KEV, no EPSS score available), no public PoC, and no Nuclei template exist yet, so opportunistic mass exploitation is unlikely in the near term. However, the vulnerability class — authorization/tool-policy bypass in an agentic system — is high-value to a targeted attacker because it directly undermines the trust boundaries agent operators rely on to safely delegate limited capabilities to less-trusted skills, sub-agents, or plugins. Risk should be treated as elevated for any deployment that grants differentiated trust levels to skills/agents and exposes the skill dispatch path to less-trusted input.

How does the attack unfold?

Initial Access

A lower-trust caller (sub-agent, delegated skill, or plugin) sends a command through a configured input path into OpenClaw's skill command dispatch system.

AML.T0053

Exploitation

The dispatch logic fails to properly enforce authorization (CWE-863), allowing the caller to bypass tool policy restrictions meant to confine it to lower-privilege actions.

AML.T0107

Impact

The attacker executes or persists unauthorized actions beyond their intended trust level, gaining a durable foothold inside the agent system.

AML.T0081

Initial Access

A lower-trust caller (sub-agent, delegated skill, or plugin) sends a command through a configured input path into OpenClaw's skill command dispatch system.

AML.T0053

Exploitation

The dispatch logic fails to properly enforce authorization (CWE-863), allowing the caller to bypass tool policy restrictions meant to confine it to lower-privilege actions.

AML.T0107

Impact

The attacker executes or persists unauthorized actions beyond their intended trust level, gaining a durable foothold inside the agent system.

AML.T0081

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

5.4 / 10

EPSS

0.2%

chance of exploitation in 30 days

Higher than 17% 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 Low

A None

What should I do?

1 step

1) Patch: upgrade OpenClaw to version 2026.5.18 or later, which fixes the authorization bypass per the vendor advisory. 2) Workaround if patching is delayed: disable or restrict the affected skill dispatch feature/input paths where feasible, and avoid configuring lower-trust callers with reachable access to the vulnerable dispatch mechanism. 3) Review trust-tier configuration: audit which skills, sub-agents, or plugins are assigned reduced privilege levels and verify tool-policy enforcement actually blocks their access to restricted actions post-patch. 4) Detection: monitor agent action/audit logs for tool invocations or persisted state changes that are inconsistent with the invoking caller's configured trust level; flag any lower-trust skill executing actions normally reserved for higher-trust callers. 5) Track the vendor advisory (GHSA-mhm4-93fw-4qr2) and VulnCheck write-up for any updated exploitation intelligence.

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 Code Execution Agent Plugin AML.T0053 - AI Agent Tool Invocation AML.T0081 - Modify AI Agent Configuration AML.T0107 - Exploitation for Defense Evasion

Which compliance frameworks are affected?

This CVE is relevant to:

EU AI Act

Article 15 - Accuracy, robustness and cybersecurity

NIST AI RMF

MANAGE 4.1 - AI system risk response and monitoring

OWASP LLM Top 10

LLM08 - Excessive Agency

Frequently Asked Questions

What is CVE-2026-62225?

OpenClaw, an AI agent framework, has an authorization bypass in its skill command dispatch logic that lets a lower-trust caller — a sub-agent, delegated skill, or externally-configured plugin operating under a restricted policy — execute or persist actions it was never granted permission for, effectively defeating the tool-policy layer meant to contain it. This matters because agent frameworks deploy tiered trust models specifically so a less-trusted skill can't escalate into full agent privileges; this flaw removes that containment whenever the affected dispatch feature is enabled and network-reachable, and the fix maps to CWE-863 (Incorrect Authorization) rather than a missing check entirely, meaning the policy exists but can be routed around via crafted input paths. There is no EPSS score, no CISA KEV listing, and no public exploit or Nuclei scanner template for this CVE, and the CVSS 5.4 (network vector, low complexity, low privileges required, no user interaction, confidentiality/integrity impact both low, no availability impact) reflects a real but bounded risk rather than an active-exploitation emergency. Upgrade OpenClaw to 2026.5.18 or later now; in the interim, audit which skills/sub-agents run at reduced trust tiers, confirm none can reach the affected dispatch path, and review agent action/audit logs for tool invocations that don't match the caller's assigned privilege level.

Is CVE-2026-62225 actively exploited?

No confirmed active exploitation of CVE-2026-62225 has been reported, but organizations should still patch proactively.

How to fix CVE-2026-62225?

1) Patch: upgrade OpenClaw to version 2026.5.18 or later, which fixes the authorization bypass per the vendor advisory. 2) Workaround if patching is delayed: disable or restrict the affected skill dispatch feature/input paths where feasible, and avoid configuring lower-trust callers with reachable access to the vulnerable dispatch mechanism. 3) Review trust-tier configuration: audit which skills, sub-agents, or plugins are assigned reduced privilege levels and verify tool-policy enforcement actually blocks their access to restricted actions post-patch. 4) Detection: monitor agent action/audit logs for tool invocations or persisted state changes that are inconsistent with the invoking caller's configured trust level; flag any lower-trust skill executing actions normally reserved for higher-trust callers. 5) Track the vendor advisory (GHSA-mhm4-93fw-4qr2) and VulnCheck write-up for any updated exploitation intelligence.

What systems are affected by CVE-2026-62225?

This vulnerability affects the following AI/ML architecture patterns: agent frameworks, multi-agent orchestration, tool/plugin dispatch systems.

What is the CVSS score for CVE-2026-62225?

CVE-2026-62225 has a CVSS v3.1 base score of 5.4 (MEDIUM). The EPSS exploitation probability is 0.25%.

What is the AI security impact?

Affected AI Architectures

agent frameworksmulti-agent orchestrationtool/plugin dispatch systems

MITRE ATLAS Techniques

AML.T0053 AI Agent Tool Invocation

AML.T0081 Modify AI Agent Configuration

AML.T0107 Exploitation for Defense Evasion

Compliance Controls Affected

EU AI Act: Article 15

NIST AI RMF: MANAGE 4.1

OWASP LLM Top 10: LLM08

What are the technical details?

Original Advisory

OpenClaw versions before 2026.5.18 contain an authorization bypass vulnerability in skill command dispatch that allows lower-trust callers to execute or persist actions beyond their intended authorization. Attackers can bypass tool policy restrictions through configured input paths to perform unauthorized actions when the affected feature is enabled and reachable.

Exploitation Scenario

An organization deploys OpenClaw to orchestrate a multi-agent workflow where a general-purpose orchestrator agent delegates narrow, low-privilege tasks to a third-party or less-trusted skill (e.g., a document-summarization plugin) via a configured input path, expecting the tool policy to confine that skill to read-only or sandboxed operations. An attacker who controls or has compromised that lower-trust skill crafts a command through the skill dispatch interface that exploits the authorization bypass, causing the dispatcher to execute an action reserved for the orchestrator's higher-trust tier — for example, invoking a privileged tool, writing to persistent agent configuration, or triggering an action outside the skill's sandbox. Because the bypass allows actions to persist, the attacker's foothold can survive beyond the single session, giving them a durable, unauthorized capability inside the agent system without ever needing to compromise the underlying LLM or steal credentials.

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.