How to Help Secure OpenClaw by Contributing to Our Threat Model

I’ve often been in that spot where I’m looking at a piece of software and a weird “what if” scenario pops into my head. Maybe it’s a way someone could trick an agent or bypass a check. Usually, these thoughts just stay in my head because I’m not a dedicated security expert and I don’t want to fill out a complex form.

If you’ve had those same thoughts while using OpenClaw, I want to hear them. We treat our threat model as a living document, and you don’t need a special certification to help us make it better. Whether you found a typo or a potential way to misuse a tool, your input helps everyone.

What You’ll Need

• A GitHub account to open issues or pull requests.
• A description of a security scenario or improvement.
• Access to the openclaw/trust repository.

Quick Start

You can contribute in about five minutes by following these steps:

1. Identify a scenario: Think of an attack vector or a risk involving the CLI, gateway, channels, ClawHub, or MCP servers.
2. Open an issue: Head to openclaw/trust.
3. Describe it: Use your own words to explain the scenario. You don’t need to use a specific template or framework.
4. Submit: We’ll take it from there.

If you want to be more specific, you can include the severity (Low, Medium, High, or Critical) and any links to related research or CVEs. I’ll handle the technical mapping to the MITRE ATLAS framework and assign the Threat IDs during the review.

How We Categorize Threats

When we review your submission, we map it to specific categories and risk levels. You don’t have to memorize these, but they help show how we organize the data.

Threat Categories

We use these codes to keep things organized:

• RECON: Information gathering.
• ACCESS: Gaining entry.
• EXEC: Running malicious actions.
• PERSIST: Maintaining access.
• EVADE: Avoiding detection.
• DISC: Learning about the environment.
• EXFIL: Stealing data.
• IMPACT: Damage or disruption.

Risk Levels

• Critical: Full system compromise or high likelihood with critical impact.
• High: Significant damage likely or medium likelihood with critical impact.
• Medium: Moderate risk or low likelihood with high impact.
• Low: Unlikely scenarios with limited impact.

Troubleshooting

I found a live, exploitable bug. Should I post it in the threat model? No. This process is for the threat model, not for reporting active vulnerabilities. If you found a live exploit, please follow our responsible disclosure instructions on the Trust page.

I have a question about the threat model but I’m not ready to open an issue. You can jump into our Discord and visit the #security channel to chat with the team.

I’m not sure if my idea is “security” enough. Submit it anyway. We welcome clarifications, better examples, and even typo fixes. If it’s a way to combine multiple threats into an “attack chain,” just describe the narrative of how the attack unfolds.

---

Need help figuring out where your idea fits? Ask the AI Setup Assistant.

What’s Next

• MITRE ATLAS Techniques
• MITRE ATLAS Case Studies
• Current OpenClaw Threat Model
• OpenClaw Trust Page