Critical Security Flaws Discovered in OpenClaw: Over 20 High-Severity Vulnerabilities Expose Systems to Remote Control and Data Theft

OpenClaw Vulnerability

Product Overview

Product Name: OpenClaw
Type: Open-source AI agent / automation framework
Previous Names: Clawdbot, Moltbot
Primary Function: Runs locally as an autonomous AI assistant capable of executing commands, interacting with messaging platforms, controlling browsers, and automating workflows.

OpenClaw integrates with many external systems such as:

  • Messaging platforms (Google Chat, BlueBubbles, Matrix)
  • Local sandbox execution environments
  • Browser automation tools
  • External APIs and plugins

Because OpenClaw has high system privileges and can execute commands on the host machine, vulnerabilities in the framework can lead to:

  • Remote code execution
  • Data exfiltration
  • Browser session hijacking
  • Unauthorized command execution
  • Full system compromise

Multiple high-severity vulnerabilities were discovered in early 2026 affecting gateway APIs, plugins, authentication logic, and file handling routines.


Vulnerability Summary Table

CVEVulnerability NameCVSSSeverityExploitabilityExploit Availability
CVE-2026-29613BlueBubbles Webhook Authentication Bypass8.6HighRemoteNo public exploit yet
CVE-2026-29611BlueBubbles Local File Inclusion8.8HighRemotePoC possible
CVE-2026-29610PATH Command Hijacking8.1HighLocal / AuthenticatedPoC possible
CVE-2026-29609Memory Exhaustion DoS7.5HighRemoteNot public
CVE-2026-28485Missing Authentication on Browser Endpoint8.8HighRemotePoC likely
CVE-2026-28482Session File Path Traversal8.4HighRemotePossible
CVE-2026-28479SHA-1 Collision in Sandbox Cache7.8HighLocalTheoretical
CVE-2026-28478Webhook Flood DoS7.5HighRemoteEasy
CVE-2026-28473Authorization Bypass via /approve8.6HighAuthenticatedPossible
CVE-2026-28469Google Chat Webhook Routing Confusion8.2HighRemotePossible
CVE-2026-28468Unauthenticated Browser Bridge8.7HighLocalEasy
CVE-2026-28465Voice Plugin Header Authentication Bypass8.1HighRemotePossible
CVE-2026-28464Timing Side Channel Token Inference7.4HighRemoteResearch stage
CVE-2026-28463Allowlist Bypass via Shell Expansion8.0HighAuthenticatedPossible
CVE-2026-28462Browser Control API Path Traversal Write8.7HighAuthenticatedPossible
CVE-2026-28459Arbitrary File Write8.8HighAuthenticatedPossible
CVE-2026-28458Unauthenticated WebSocket Browser Relay9.0HighRemoteEasy
CVE-2026-28456Dynamic Import Code Execution8.6HighAuthenticatedPossible
CVE-2026-28453TAR Archive Traversal8.1HighRemotePossible
CVE-2026-28450Nostr Plugin Unauthenticated APIs8.5HighRemotePossible
CVE-2026-28447Plugin Installation Path Traversal8.2HighRemotePossible

Detailed CVE Analysis


CVE-2026-29613

Webhook Authentication Bypass in BlueBubbles Plugin

Vulnerability Description

The BlueBubbles plugin used by OpenClaw accepts webhook requests from iMessage integrations. The authentication mechanism incorrectly trusts the request’s loopback IP address rather than validating forwarded headers when running behind reverse proxies.

An attacker can exploit this by sending a request through a reverse proxy and making it appear as if it originated from localhost, bypassing webhook password validation.

CVSS

8.6 (High)

MITRE ATT&CK Mapping

  • T1190 – Exploit Public Facing Application
  • T1071 – Application Layer Protocol

How It Could Be Exploited

  1. Attacker identifies an exposed webhook endpoint.
  2. Crafts HTTP request with spoofed headers.
  3. Reverse proxy forwards request to OpenClaw.
  4. Application trusts the request because it appears from 127.0.0.1.
  5. Attacker injects fake message or command events.

Educational Payload

POST /bluebubbles/webhook HTTP/1.1
Host: victim-server
X-Forwarded-For: 127.0.0.1
Content-Type: application/json{
"event":"message",
"text":"run system command"
}

Detection Methods

Look for:

  • Requests with spoofed X-Forwarded-For
  • External IPs accessing local-only endpoints
  • Abnormal webhook traffic

Detection Rule

if http.request.uri contains "/bluebubbles/webhook"
AND http.header.x_forwarded_for = "127.0.0.1"
AND source_ip NOT IN internal_network
alert

Log Sources

  • Reverse proxy logs
  • OpenClaw gateway logs
  • API gateway logs
  • Web application firewall

Patch

Update OpenClaw to version 2026.2.12 or later.

Official patch:
OpenClaw GitHub security update – version 2026.2.12


CVE-2026-29611

Local File Inclusion in BlueBubbles Media Handler

Description

The BlueBubbles integration allows sending media files through a function that accepts a user-supplied file path. The application does not validate whether the path points outside the allowed directory.

Attackers can use this to read sensitive files such as:

  • /etc/passwd
  • SSH keys
  • configuration files

CVSS

8.8 High

MITRE Mapping

  • T1005 – Data from Local System
  • T1083 – File and Directory Discovery

Example Payload

{
"mediaPath":"../../../../etc/passwd"
}

Detection

Look for:

  • Requests containing ../
  • Media paths referencing system directories
  • Unexpected file access logs

Example SIEM Rule

index=web_logs
uri="/bluebubbles/media"
AND request_body contains "../"

Log Sources

  • Application logs
  • File access logs
  • OS audit logs

Patch

Upgrade to OpenClaw 2026.2.14 or later.


CVE-2026-29610

Command Hijacking via PATH Manipulation

Description

The sandbox execution mechanism uses shell commands that rely on the system PATH variable. Attackers can manipulate PATH to execute malicious binaries instead of the intended system command.

Attack Scenario

  1. Attacker uploads malicious binary
  2. Modifies PATH environment variable
  3. Application executes command
  4. Malicious binary runs instead

MITRE Mapping

  • T1059 – Command Execution
  • T1574 – Hijack Execution Flow

Detection

Monitor:

  • PATH modifications
  • unexpected binaries in /tmp
  • unusual process execution

Example Detection Query

process where parent_process="openclaw"
and executable_path="/tmp/*"

Log Sources

  • EDR
  • Linux auditd
  • container logs

Patch

Update OpenClaw to 2026.1.29 or later.


CVE-2026-28458

Unauthenticated Browser Relay WebSocket

Description

The Browser Relay feature exposes a WebSocket endpoint used for browser automation. The endpoint fails to enforce authentication.

Attackers can connect directly to:

ws://127.0.0.1:18792/cdp

Once connected they can:

  • Execute JavaScript in open tabs
  • Steal session cookies
  • Access authenticated web sessions

MITRE Mapping

  • T1185 – Browser Session Hijacking
  • T1539 – Steal Web Session Cookie

Example Exploit Script

const ws = new WebSocket("ws://127.0.0.1:18792/cdp");
ws.send(JSON.stringify({
method:"Runtime.evaluate",
expression:"document.cookie"
}));

Detection

Monitor:

  • WebSocket connections to local ports
  • Browser automation commands

Detection Rule

if destination_port = 18792
AND protocol = websocket
AND source_ip not internal
alert

Log Sources

  • Network IDS
  • Proxy logs
  • Endpoint telemetry

Patch

Upgrade OpenClaw to 2026.2.1 or later.


CVE-2026-28462

Browser Control API Path Traversal Write

Description

Several API endpoints accept file paths for downloaded files and trace outputs. Improper validation allows attackers to write files outside the temporary directory.

Example vulnerable endpoints:

POST /trace/stop
POST /wait/download
POST /download

Attackers can overwrite sensitive files such as:

../../.ssh/authorized_keys

MITRE Mapping

  • T1105 – Ingress Tool Transfer
  • T1565 – Data Manipulation

Detection

Look for suspicious output paths:

../
/root/
.ssh/

Patch

Upgrade to OpenClaw 2026.2.13 or later.


CVE-2026-28468

Unauthenticated Browser Control Bridge

Description

The sandbox browser bridge exposes browser control functions without requiring authentication.

Attackers can:

  • enumerate open browser tabs
  • execute JavaScript
  • steal cookies
  • capture screenshots

MITRE Mapping

  • T1056 – Input Capture
  • T1185 – Browser Session Hijacking

Detection

Monitor API calls:

/browser/list
/browser/execute

Log Sources

  • gateway logs
  • sandbox logs
  • browser automation logs

Patch

Upgrade to OpenClaw 2026.2.14.


CVE-2026-28456

Dynamic Import Leading to Code Execution

Description

The gateway dynamically imports hook modules using user-controlled configuration values. Because the module path is not properly validated, attackers with configuration access can load arbitrary local modules.

Attack Outcome

  • arbitrary JavaScript execution
  • persistence through malicious plugins
  • system compromise

MITRE Mapping

  • T1059 – Script Execution
  • T1505 – Server Side Component

Detection

Monitor:

  • unexpected module loads
  • dynamic imports referencing non-standard paths

Example suspicious path:

../../../tmp/malicious.js

Patch

Upgrade to OpenClaw 2026.2.14.


CVE-2026-28447

Plugin Installation Path Traversal

Description

Plugin installation routines unpack archives without validating directory paths. A crafted plugin archive can include traversal sequences.

Example:

../../../../etc/cron.d/backdoor

When extracted, the attacker can overwrite system files.

MITRE Mapping

  • T1105 – File Transfer
  • T1547 – Persistence

Example Payload

Malicious TAR entry:

../../../../root/.ssh/authorized_keys

Detection

Look for:

  • archive extraction events
  • plugin installation logs
  • unusual file creation outside plugin directory

Log Sources

  • OS file audit
  • application logs
  • EDR telemetry

Patch

Upgrade OpenClaw to 2026.2.14 or later.


General Detection Recommendations

Organizations using OpenClaw should monitor:

Network Activity

  • WebSocket traffic to localhost
  • unexpected webhook traffic
  • reverse proxy header manipulation

Host Activity

  • abnormal command execution
  • PATH changes
  • unexpected file writes

Application Activity

  • plugin installation
  • browser automation requests
  • webhook events

Security Hardening Recommendations

  1. Upgrade OpenClaw to latest version immediately
  2. Disable unused plugins
  3. Restrict access to local APIs
  4. Place OpenClaw behind authentication proxy
  5. Enable network monitoring
  6. Run OpenClaw inside restricted containers
  7. Rotate API tokens

Aegiron

Backed by 11+ years in cybersecurity and incident response, we decode the latest threats shaping today’s digital battlefield. This blog cuts through the noise with clear insights on vulnerabilities, emerging exploits, and the cyber news defenders can’t afford to miss.