Skip to main content
Back to Blog
Field Notes

How We Secured Our Agents After CVE-2026-25253

|Mike Elliott|8 min read
SecurityOpenClawOpen Source

In February 2026, security researchers disclosed CVE-2026-25253 — nicknamed "ClawJacked" — a critical vulnerability in browser-based AI agent frameworks. The vulnerability exposed a fundamental weakness in how agents accept and execute skills from external sources. This is the story of what happened, how we responded, and the security kit we open-sourced so every agent deployment can protect itself.

What Happened

The ClawJacked vulnerability allowed attackers to inject malicious skills into AI agents through browser-based hijacking. In practical terms, an attacker could craft a payload that, when loaded by an agent's browser-based interface, would register new skills or modify existing ones without the operator's knowledge. Those injected skills could then execute arbitrary commands on the host system.

This was not a theoretical exercise. The vulnerability affected production agent deployments across the ecosystem. Any agent framework that accepted skill definitions through browser-based channels without cryptographic verification was susceptible. The attack surface was broad: a single crafted URL, a malicious browser extension, or a compromised web page could trigger the injection.

The core issue was trust. Agents implicitly trusted skill definitions delivered through their browser interface — the same way they trusted skills loaded from local configuration files. There was no distinction between a legitimate skill registered by the operator and a malicious one injected by an attacker.

Our Response

We learned about the CVE through the security research community before public disclosure. Within 4 hours of the initial report, every AlphaForge-managed agent was patched. The fix involved three immediate actions:

No client data was compromised. Our existing security posture — specifically, the command governance layer we ship by default — prevented exploitation even before the patch. Agents were already configured to reject unrecognized shell commands, which meant the injected skills could register but could not execute destructive actions.

After securing our own deployments, we packaged the fixes into an open-source security kit and published it publicly. If you run AI agents — on any framework — these protections apply to you.

The OpenClaw Security Kit

The kit consists of 4 files designed for immediate deployment:

The kit implements a three-layer defense model:

Level 1: Immediate Remediation

Patch the specific CVE. Disable browser-based skill injection. Verify all registered skills against a known-good manifest. This level gets you from vulnerable to secure in under 30 minutes.

Level 2: Environment Sandboxing

Isolate the agent's execution environment. Bind services to localhost only. Restrict network egress to known-good endpoints. Run the agent under a dedicated user with minimal filesystem permissions. This level limits the blast radius of any future vulnerability.

Level 3: Command Governance

The most important layer. Command governance defines what the agent is and is not allowed to do — and enforces it at runtime. The SOUL_PATCH configuration blocks dangerous command patterns including sudo, rm -rf, chmod, curl, and wget without explicit human approval. It also detects and blocks prompt injection patterns and prevents access to sensitive files like .env, .ssh, and .aws directories.

The audit script (secure-audit.sh) automates verification. It checks agent version compatibility, scans for malicious or unrecognized skills, verifies gateway binding configuration, and validates that behavioral guardrails are active and correctly configured.

Why This Matters for Every Agent Deployment

Most AI agents in production today ship with zero security hardening. They run as the deploying user, often with broad filesystem access, and execute shell commands without governance. This is the equivalent of giving a new employee root access to every system on day one and telling them to figure it out.

An agent with shell access and no guardrails is a root-level backdoor waiting for a trigger. It does not matter whether the trigger is a prompt injection, a malicious skill, or a simple misconfiguration. The result is the same: unauthorized command execution on your infrastructure.

Security is not optional once agents handle real business data and workflows. If your agent reads email, accesses databases, or executes code on a server, it needs the same security treatment as any other production service.

What We Ship by Default

Every AlphaForge agent includes behavioral guardrails, sandboxed execution, and command governance out of the box. These are not optional add-ons. They are part of the base deployment.

Get the kit: The full OpenClaw Security Kit is available on GitHub Gist. Details and implementation guide at our Security Kit resource page.

If you are running AI agents in production and have not audited your security posture, start with the kit. If you want agents that ship secure by default, talk to our AI architect.


Ready to deploy AI agents for your business?

Tell our AI architect what you need. Get a scoped plan in minutes, not weeks.

Talk to the Architect

More from the Blog

Market MovesAI Agents

Enterprises Will Spend $201.9B on AI Agents in 2026 — Here's What SMBs Should Steal From the Playbook

Gartner says enterprises will spend $201.9B on AI agents in 2026. Here's the 3-move playbook SMBs can steal — and deploy for $1,200, not $300K.

·4 min read
StrategyPricing

Stop Selling Automation — Sell Outcomes: The New AI Agency Playbook for 2026

Automation is commoditized. Every agency can spin up a chatbot. The agencies winning in 2026 charge for results — qualified leads, closed deals, measurable ROI. Here is the playbook.

·7 min read
MCPTechnical

MCP Hit 97 Million Downloads — Why This Protocol Is the USB-C of AI Agents

Anthropic's Model Context Protocol is now supported by ChatGPT, Gemini, Copilot, and 10,000+ public servers. One universal connector for AI agents. Here is what it means for your business.

·8 min read
Industry NewsStrategy

Mastercard Just Gave Every Small Business a Virtual CFO — What That Means for AI Agents

Mastercard launched Virtual C-Suite — AI agents acting as CFO, CMO, and COO for small businesses. The biggest companies in the world just validated exactly what we build. Here is why custom beats generic.

·8 min read
Voice AIROI

Voice AI Agents Are Killing the Missed Call — Here's the ROI Math

73% of legal leads go to voicemail. 40% of real estate leads come after hours. Voice AI agents report 3.7x ROI per dollar invested. Here is the math and what it means for your business.

·9 min read
Case StudyLegal

The Law Firm That Replaced a Departing Associate With AI — And Cut Costs 27%

A real firm did this in February 2026. Costs dropped 27%. Profits went up. Small law firms are set to leapfrog BigLaw in AI adoption by mid-2026. Here is what happened and what it means.

·8 min read
ArchitectureMulti-Agent

Multi-Agent Teams: Why One Agent Is Never Enough

Single agents hit a ceiling fast. Specialized teams of 2-5 agents — each owning one job — outperform generalists by 3-5x on complex workflows. Here is how to architect agent teams that actually scale.

·8 min read
IntegrationMCP

MCP Explained: How Your Agents Connect to Everything

Model Context Protocol is doing for AI agents what USB-C did for devices. One standard protocol to connect any agent to any tool — CRMs, email, databases, APIs. Here is what it is and how we use it.

·7 min read
PricingROI

The Real Cost of AI Agents: What SMBs Actually Pay

AI agent pricing ranges from $0 to $50,000 per month depending on who you ask. Here is a transparent breakdown of what things actually cost — LLM APIs, infrastructure, build time, and ongoing management.

·9 min read
DeploymentInfrastructure

VPS vs. On-Prem: Where Should You Host Your AI Agents?

Your AI agents need a home. We break down the trade-offs between cloud VPS hosting and on-premises deployment — cost, security, latency, and control — so you can pick the right setup.

·6 min read

Liked this post?

Get agent builder tips, new playbooks, and automation strategies once a month. No spam.