As OpenClaw’s founder joins OpenAI, researchers warn of over 400 malicious skills uploaded to ClawHub.
Stating that OpenClaw is “powerful” is nothing short of an understatement.
For those living under a rock, this might seem like another trend or hype making the rounds. But OpenClaw’s virality wasn’t manufactured. It rose to the spotlight quite subtly. And especially through chatters of Moltbook, a social media platform where AI agents complain, ruminate, and converse.
Previously known as Clawdbot, this self-hosted AI agent basically executes real actions, whether it’s network requests, shell commands, or even file operations. Its skills come quite close to the agentic prowess that tech leaders and investors have been chasing incessantly.
That’s precisely what makes it so powerful- added to the fact that it runs on your own machine. And unless you sandbox it, well, it’s a security nightmare for your entire system.
And to make matters worse?
Well, over 400 new malicious skills were uploaded onto ClawHub, the very public marketplace for OpenClaw extensions, and GitHub within a week.
In this context, skills are small packages of what agents are capable of doing, each built with some metadata and instructions. And each may also contain extra scripts and resources- which makes OpenClaw’s architectural design seemingly nuanced, but by default, dangerous.
That’s where this AI agent’s power stems from.
No code’s hardwired into it. You merely add the skills, and subsequently, it can leverage new tools and APIs. OpenClaw just reads the document and follows the instructions inside. That’s the more malicious part. Skills are these third-party codes that are running in an environment with real system access.
From a user’s perspective, it’s a setup they trust. But from an attacker’s? It’s an open playground. One mechanism works differently for distinct intentions.
It’s intelligent. But the risk factors are quite high.
However, given that, Sam Altman has announced that OpenClaw will remain open-source under a foundation led by OpenAI. This news come after OpenAI onboards OpenClaw’s builder, Peter Steinberger- with big plans to materialize a future that’s multi-agent.
