Open source AI agent OpenClaw, formerly known as Clawdbot and Moltbot, has taken over the internet by storm. The tool allows practically anybody to create autonomous AI agents that can complete complex tasks on your computer, like browsing the web and running scripts.

It’s a powerful new take on AI that comes with inherent dangers. After all, you’re letting an AI model loose on your machine, going far outside the confines of the browser-based chatbots we’ve grown accustomed to. What could possibly go wrong?

OpenClaw has caught on like wildfire, including in China, as Bloomberg reports, with users on social media bragging about “raising lobsters,” a nod to OpenClaw’s red crustacean mascot. Even tech giants like Tencent and Alibaba are adopting the tech for their own software, and government agencies are signing contracts with startups that are also leveraging OpenClaw tech.

Meetups of the OpenClaw obsessed in the country are “beginning to border on the cult-like,” Bloomberg‘s Zheping Huang wrote. “A gathering in Shenzhen this past weekend featured a room of attendees wearing tall cartoon hats resembling cooked red lobsters.”

But given the considerable cybersecurity implications, it’s no wonder Chinese authorities are starting to crack down on the phenomenon.

As Reuters reports, government agencies and state-owned enterprises are warning their staff not to install OpenClaw agents on their devices, citing security reasons, including leaks, the mistaken deletion of data, and the misuse of sensitive information.

One inside source told Reuters that OpenClaw wasn’t banned outright at the government agency they worked at, but that staff had been discouraged from installing it.

It’s an intriguing development, especially considering the sheer amount of enthusiasm for OpenClaw and AI overall inside the country, with Beijing promoting a wide-reaching AI-positive action plan to grow its economy.

But given some of the horror stories we’ve come across, the warnings are certainly warranted, especially when it comes to the handling of sensitive government data.

For instance, an executive at Meta — a company that also banned employees from using OpenClaw on their work machines — watched helplessly as a bot started wiping her “important” emails.

“Nothing humbles you like telling your OpenClaw ‘confirm before action’ and watching it speedrun deleting your inbox,“ Summer Yue, the director of safety and alignment at Meta’s Superintelligence lab, tweeted last month.

More on OpenClaw-like agents: New AI Agent Logs Directly Into College Platform Canvas to Do Your Homework for You