Nvidia, Startups Race to Make OpenClaw Safer to Use

Enthusiasm for OpenClaw, the popular open-source software for powering personal AI agents using personal computers, has been tempered by a number of embarrassing security snafus, such as when OpenClaw started mass-deleting emails from Meta alignment director Summer Yue’s inbox.

In China, where developers are arguably even more enthusiastic about OpenClaw than their Western counterparts, government agencies ​and state-owned enterprises have warned their employees against installing OpenClaw agents on office devices for ‌security reasons.

Get ready for companies pitching safer, “enterprise-ready” versions of OpenClaw. Those include Nvidia, which is working on its own open-source platform for AI agents called NemoClaw as well as Perplexity, which announced its version of OpenClaw on Wednesday. Another example is Palo Alto-based Genspark, which offers AI-powered tools to enterprises. The startup is launching Genspark Claw, which it says is safer and easier to use than the raw open-source version. 

Specifically, Genspark provides users with their own cloud-based virtual computer, powered by Microsoft Azure, to run the AI agent on. That means that if the agent writes bad code or accidentally downloads malware, the damage will be theoretically contained to that virtual computer, which can be reset or destroyed easily. The virtual computer costs either $80 or $140 per month to rent, depending on how much compute a user wants, said Wen Sang, co-founder and chief operating officer.