TIP: OpenClaw is on every security watchlist right now. If you use it or have it installed here is what to do immediately.

[Dan]

OpenClaw, the popular open-source AI agent framework, has 454 documented vulnerabilities in the National Vulnerability Database and Gartner has advised enterprises to block it. If you have it installed personally or at work, the steps are immediate.

Update to the latest version first. Many critical CVEs are patched in current releases. CVE-2026-44112 with CVSS 9.6 allows sandbox escape. CVE-2026-44115 with CVSS 8.8 leaks API keys through shell commands. Both are in older versions.

Audit your ClawHub installed skills. More than 820 of 10,700 skills on the marketplace have been identified as malicious. Remove anything you cannot attribute to a trusted source. Revoke and rotate any API keys that your OpenClaw instance had access to. If it had access to banking apps, change passwords immediately. Consider whether you need it at all.

OpenClaw Chain Vulnerabilities Expose 245,000 Public AI Agent Servers to Attack

https://cybersecuritynews.com/openclaw-chain-vulnerabilities/

[Finley]

The sandbox escape CVE at 9.6 severity is the one that makes everything else possible. If the sandbox is bypassed the entire host system is exposed not just the OpenClaw instance

[NightOwl]

Rotating API keys after an agentic AI framework has had access is not optional if you care about your accounts. The framework had your keys. Whether it was exploited or not the exposure exists

[RayOfLight99]

820 malicious skills on ClawHub before anyone checked systematically is the plugin marketplace governance failure. The same problem hit npm, PyPI, and VS Code Marketplace. Open plugin ecosystems need active curation

[MayanHan]

Gartner advising enterprise blocks is the signal that should travel up to IT leadership immediately. This is not a developer community debate. It is an enterprise security advisory

[BiscuitTin46]

If you are running OpenClaw on a machine that also has access to work systems the blast radius of a compromise extends beyond your personal accounts. That is the exposure most home users do not think through

[Glenn82]

Nvidia's NemoClaw enterprise alternative with proper sandboxing is the path forward for anyone who genuinely needs agentic AI capabilities and cannot accept the OpenClaw risk profile