OpenClaw has launched a public security hub that gathers its advisories, vulnerability-reporting process and ongoing hardening work in one place. The open-source AI project says it has published 722 fixes since January, including fixes for 14 reports classified as critical vulnerabilities.
OpenClaw Watch analysis
What happened
In a 16 September update, the OpenClaw Security Team said the new page brings together material previously spread across GitHub, documentation, blog posts and individual advisories. It includes current security status, active bulletins, guidance on submitting vulnerability reports and an explanation of what falls outside the project's security model.
The project says its security work covers gateway authentication and scopes, sandboxing, approvals, connectors, filesystem access, plugins, skills and network access. It also lists ongoing measures including ClawHub skill scanning, blocking installs of malicious or quarantined skills, command-chain checks, network-egress controls and regression testing based on previously fixed advisories.
OpenClaw also says it has published an open dataset containing more than 67,000 ClawHub skill scans, alongside research into how different security scanners assess the same skills. The project says potential vulnerabilities should be reported privately first, giving maintainers time to investigate and ship a fix before technical details are made public. Explore OpenClaw's security updates.
Why it matters
OpenClaw connects AI models to tools, files, browsers and external services, so a flaw can sit at the boundary between an agent's instructions and the systems it can actually touch. A single public hub does not make those boundaries safe, but it makes the project's security posture easier to inspect and gives researchers a clearer route to report problems.
The useful reader takeaway is practical: check the security page and advisories before installing skills or connecting OpenClaw to sensitive systems, and use the private reporting route if you find a suspected vulnerability. The headline numbers remain figures supplied by OpenClaw, rather than an independent audit. Still, publishing the underlying scanning dataset is a more useful move than asking everyone to admire the word “security” from a distance.
Our read
This is a worthwhile security release because it turns scattered maintenance work into something users and researchers can actually consult. The strongest feature is not the 722-fix tally, impressive though it is. It is the combination of status information, reporting guidance, disclosed fixes and an open dataset.
The next test is whether the hub stays current, links clearly to resolved advisories and makes it possible to understand the limits of each security measure. OpenClaw is building an agent platform with a large attack surface. Showing the plumbing is not glamorous, but it is how trust begins.
What to watch
- Whether the security hub records active advisories and newly fixed issues promptly.
- Whether the 67,000-scan dataset receives documentation and useful updates.
- How effectively ClawHub scanning and install blocking catch malicious or risky skills.
- Whether independent researchers can reproduce or challenge the project's security claims.
Discussion spark: Should open-source AI projects be expected to publish security datasets and detailed fix histories as standard, or could that transparency expose too much about systems that are still under active attack?
Sources and evidence
- Source update (16 September 2026, 00:00 UTC)
OpenClaw Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by the OpenClaw Foundation.