SecurityScorecard published a deep analysis of internet-exposed OpenClaw deployments this week, and the numbers are striking. While the media coverage focused on a malware campaign that borrowed the OpenClaw name, the more important finding is about legitimate OpenClaw installations that operators left exposed without protective measures.

Here's what the research actually says — and what it means for your setup.

The Numbers

40,214
Internet-exposed OpenClaw instances found
28,663
Unique IPs with accessible control panels
63%
Vulnerable to remote code execution
549
Linked to previous breach activity
1,493
Tied to known active vulnerabilities
7.8–8.8
CVSS scores on the three high-severity CVEs

The three CVEs with CVSS scores between 7.8 and 8.8 all have public exploit code available. That means anyone can download a working exploit and run it against an exposed instance — no sophistication required.

What "Exposed" Actually Means

When SecurityScorecard says "internet-exposed," they mean the OpenClaw gateway control panel is reachable from the public internet — no VPN, no authentication layer in front, just a direct connection from anywhere in the world.

This happens for a few common reasons:

The problem isn't running OpenClaw. It's running OpenClaw with the control panel accessible to the entire internet.

The name identification risk: SecurityScorecard also noted that users commonly configure their OpenClaw bots with identifiable personal or company names — making them easily findable targets. If your gateway responds to requests with your name or business name in the headers or UI, attackers can use Shodan or similar tools to find you specifically.

What an Attacker Can Do with a Vulnerable Exposed Instance

The three CVEs in the research enable remote code execution — meaning an attacker who reaches your exposed gateway can run arbitrary commands on the machine hosting it. With a CVSS of 7.8-8.8, that's high-severity without requiring any authentication.

From there, the attacker has everything OpenClaw has access to:

"In practice, because it was written by AI, security wasn't a dominating feature in the development process." — Jeremy Turner, VP of Threat Intelligence, SecurityScorecard

Turner's quote is worth sitting with. It applies to the malware campaign they studied — but it's also a fair characterization of many self-hosted OpenClaw setups. People move fast, follow convenience-first tutorials, and don't think about the exposure surface until something goes wrong.

The Fix: Three Config Changes That Remove You from the Risk Pool

The good news is that fixing this is not hard. The 40,214 exposed instances in SecurityScorecard's scan share a common trait: their gateways are reachable from the internet. Remove that exposure and you remove the attack surface for the CVEs entirely.

If you're on ClawReady's setup: Your gateway was configured with bind: 127.0.0.1 from day one. You're not in the 40,214. If you've made changes since your initial setup, run openclaw doctor to verify your current config state.

Why the Scale Is So Large

40,000+ exposed instances is a lot — but it reflects the pace of OpenClaw adoption in 2026. The framework went from niche to mainstream extremely quickly. A huge wave of first-time self-hosters followed tutorials written for convenience, deployed to cloud VMs, and left their gateways public-facing.

This pattern repeats with every fast-growing open-source project. It happened with Jupyter notebooks (thousands exposed with no auth), with MongoDB (hundreds of millions of records exposed), with Elasticsearch. The attack surface scales with adoption, and the attackers know it.

OpenClaw is now mainstream enough to be worth scanning for at scale. That's not a reason to avoid it — it's a reason to configure it properly.

The SecurityScorecard Research vs. the "OpenClaw Trojan" Story

Earlier this week, Tweaktown covered a malware campaign called the "OpenClaw Trojan" — malware that borrowed the OpenClaw name to appear legitimate (we covered it here). That's a separate issue from what SecurityScorecard found.

SecurityScorecard's research is about legitimate OpenClaw installations that are misconfigured and exposed. The trojan story is about a different piece of software that borrowed the name. Both are real security concerns — they just affect different populations.

If you downloaded OpenClaw from npm or GitHub, you have the real thing. The question SecurityScorecard is raising is: is your real OpenClaw exposed to the internet, and is it patched?

Quick Self-Audit

Take 5 minutes right now:

  1. Run openclaw --version — compare against the latest on npmjs.com/package/openclaw
  2. Check your openclaw.json for gateway.bind — it should be 127.0.0.1, not 0.0.0.0
  3. If you're on a VPS, run ss -tlnp | grep openclaw — if the gateway port shows 0.0.0.0, you're exposed
  4. Run openclaw doctor — check for any config warnings

If any of those checks come back wrong, fix them before you do anything else today.

Not sure if your setup is exposed?

We'll audit your gateway binding, version, config, and network exposure. You'll get a plain-English report of what's at risk and a prioritized fix list. Takes 24 hours or less.

Book a Security Audit — $49

Summary

SecurityScorecard found 40,214 internet-exposed OpenClaw instances, with 63% vulnerable to remote code execution via CVEs with CVSS scores up to 8.8 and publicly available exploit code. This is a configuration problem, not a framework problem — operators who left gateways public-facing without patching are at risk. Operators who bind to localhost, use VPN for remote access, and stay updated are not in the risk pool.

Three changes: bind to 127.0.0.1, add Tailscale for remote access, update and run openclaw doctor --fix. Do those and you're out of the 40,000.