CVE-2026-32028: OpenClaw auth bypass enables

CISO Take

OpenClaw's AI agent framework fails to enforce dmPolicy and allowFrom authorization checks on Discord DM reaction events, allowing any unauthenticated network attacker to enqueue reaction-derived system events and trigger downstream agent automation and tool policies. Although CVSS scores this at medium (5.3), EPSS places exploitation probability in the top 90th percentile—a meaningful signal for teams running Discord-connected AI agents with broad tool access. No public exploits exist and only 4 downstream packages depend on OpenClaw, keeping blast radius contained for now, but the pattern of bypassing agent activation controls via unvalidated event sources mirrors broader AI agent authorization anti-patterns documented in AIID incident #1368, where the same OpenClaw ecosystem was abused to deliver credential-stealing malware. Upgrade to OpenClaw 2026.2.25 or later immediately; if patching cannot be done immediately, disable Discord DM reaction triggers or enforce allowlist restrictions at the Discord platform level.

Sources: NVD EPSS GitHub Advisory VulnCheck ATLAS

What is the risk?

Medium operational risk elevated by EPSS top-90th-percentile signal. The network-accessible attack vector with no authentication required and low complexity makes this trivially easy to attempt, though CVSS-measured impact is limited to integrity (I:L). The real risk amplification lies in the AI agent context: triggering downstream automation or tool policies can cascade into privileged actions—file writes, API calls, or lateral movement—well beyond what CVSS captures depending on which tools the agent has configured. Organizations running OpenClaw agents with permissive tool sets should treat this as higher operational risk than the base score implies. The package's history of 396 CVEs further suggests a weak security posture warranting heightened scrutiny.

How does the attack unfold?

Target Identification

Attacker searches public Discord server listings, GitHub repositories, or ClawHub to identify an OpenClaw-powered bot operating in DM channels with reaction-based triggers enabled.

AML.T0095

Authorization Bypass

Attacker reacts to any bot-authored DM message using a standard emoji; vulnerable OpenClaw versions fail to validate the reactor identity against dmPolicy/allowFrom, accepting the event as authorized.

AML.T0049

Event Queue Injection

The unvalidated reaction is enqueued as a legitimate reaction-derived system event, bypassing the intended authorization layer and entering the agent's downstream processing pipeline.

AML.T0084.002

Unauthorized Tool Execution

Downstream tool policies and automation handlers process the injected event, executing privileged operations—file access, API calls, or code execution—on behalf of the unauthorized user.

AML.T0053

Target Identification

Attacker searches public Discord server listings, GitHub repositories, or ClawHub to identify an OpenClaw-powered bot operating in DM channels with reaction-based triggers enabled.

AML.T0095

Authorization Bypass

Attacker reacts to any bot-authored DM message using a standard emoji; vulnerable OpenClaw versions fail to validate the reactor identity against dmPolicy/allowFrom, accepting the event as authorized.

AML.T0049

Event Queue Injection

The unvalidated reaction is enqueued as a legitimate reaction-derived system event, bypassing the intended authorization layer and entering the agent's downstream processing pipeline.

AML.T0084.002

Unauthorized Tool Execution

Downstream tool policies and automation handlers process the injected event, executing privileged operations—file access, API calls, or code execution—on behalf of the unauthorized user.

AML.T0053

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

5.3 / 10

EPSS

0.2%

chance of exploitation in 30 days

Higher than 10% 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 None

I Low

A None

What should I do?

4 steps

Patch: Upgrade OpenClaw to version 2026.2.25 or later per vendor advisory GHSA-354r-7mfh-7rh2 and the linked commit aedf62ac.
Workaround if immediate patching is not possible: Disable Discord DM reaction event processing in agent configuration, or restrict bot-authored DM channels to allowlisted users at the Discord application permission level.
Detection: Audit Discord bot event logs for reaction events originating from users not in the dmPolicy/allowFrom list; flag any reaction-triggered automation runs from unapproved users.
Scope reduction: Review which tool policies and automations are accessible via reaction-derived events and apply least-privilege scoping to limit what any reaction event can trigger.

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 Agent Plugin AML.T0049 - Exploit Public-Facing Application AML.T0053 - AI Agent Tool Invocation AML.T0084.002 - Activation Triggers

Which compliance frameworks are affected?

This CVE is relevant to:

EU AI Act

Article 9 - Risk management system

ISO 42001

A.6.2.2 - Access control for AI systems

NIST AI RMF

MANAGE 2.2 - AI risk management plans include process improvements and risk treatment

OWASP LLM Top 10

LLM07 - Insecure Plugin Design LLM08 - Excessive Agency

Frequently Asked Questions

What is CVE-2026-32028?

OpenClaw's AI agent framework fails to enforce dmPolicy and allowFrom authorization checks on Discord DM reaction events, allowing any unauthenticated network attacker to enqueue reaction-derived system events and trigger downstream agent automation and tool policies. Although CVSS scores this at medium (5.3), EPSS places exploitation probability in the top 90th percentile—a meaningful signal for teams running Discord-connected AI agents with broad tool access. No public exploits exist and only 4 downstream packages depend on OpenClaw, keeping blast radius contained for now, but the pattern of bypassing agent activation controls via unvalidated event sources mirrors broader AI agent authorization anti-patterns documented in AIID incident #1368, where the same OpenClaw ecosystem was abused to deliver credential-stealing malware. Upgrade to OpenClaw 2026.2.25 or later immediately; if patching cannot be done immediately, disable Discord DM reaction triggers or enforce allowlist restrictions at the Discord platform level.

Is CVE-2026-32028 actively exploited?

No confirmed active exploitation of CVE-2026-32028 has been reported, but organizations should still patch proactively.

How to fix CVE-2026-32028?

1. Patch: Upgrade OpenClaw to version 2026.2.25 or later per vendor advisory GHSA-354r-7mfh-7rh2 and the linked commit aedf62ac. 2. Workaround if immediate patching is not possible: Disable Discord DM reaction event processing in agent configuration, or restrict bot-authored DM channels to allowlisted users at the Discord application permission level. 3. Detection: Audit Discord bot event logs for reaction events originating from users not in the dmPolicy/allowFrom list; flag any reaction-triggered automation runs from unapproved users. 4. Scope reduction: Review which tool policies and automations are accessible via reaction-derived events and apply least-privilege scoping to limit what any reaction event can trigger.

What systems are affected by CVE-2026-32028?

This vulnerability affects the following AI/ML architecture patterns: agent frameworks, Discord bot integrations, event-driven automation pipelines.

What is the CVSS score for CVE-2026-32028?

CVE-2026-32028 has a CVSS v3.1 base score of 5.3 (MEDIUM). The EPSS exploitation probability is 0.20%.

What is the AI security impact?

Affected AI Architectures

agent frameworksDiscord bot integrationsevent-driven automation pipelines

MITRE ATLAS Techniques

AML.T0049 Exploit Public-Facing Application

AML.T0053 AI Agent Tool Invocation

AML.T0084.002 Activation Triggers

Compliance Controls Affected

EU AI Act: Article 9

ISO 42001: A.6.2.2

NIST AI RMF: MANAGE 2.2

OWASP LLM Top 10: LLM07, LLM08

What are the technical details?

Original Advisory

OpenClaw versions prior to 2026.2.25 fail to enforce dmPolicy and allowFrom authorization checks on Discord direct-message reaction notifications, allowing non-allowlisted users to enqueue reaction-derived system events. Attackers can exploit this inconsistency by reacting to bot-authored DM messages to bypass DM authorization restrictions and trigger downstream automation or tool policies.

Exploitation Scenario

An attacker discovers an OpenClaw-powered Discord bot operating in a DM channel—identified via public Discord server listings, GitHub repository searches, or by being invited to the bot's DM scope. The attacker reacts to any message authored by the bot using a standard emoji reaction. Vulnerable OpenClaw versions fail to validate the reactor identity against dmPolicy or allowFrom, treating the reaction as a legitimate activation trigger. This enqueues a reaction-derived system event that bypasses the authorization layer entirely, causing the agent to process it as if it came from an authorized user. Depending on configured tool policies, the attacker can trigger automated workflows including data retrieval, external API calls, or code execution—co-opting the agent's privileged tool access without any credentials.

Weaknesses (CWE)

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.