Following the Trend of "Raising Lobsters"? Be careful not to get "pinched" by the shrimp!

The AI craze sparked by the robot showcasing martial arts on the Spring Festival Gala has not yet faded, and “Lobster Farming” has become a new trend. But this “lobster” isn’t seafood; it’s a highly capable AI agent—OpenClaw.

Why is everyone eager to “farm lobsters”? Wouldn’t a powerful “lobster” installed on your phone or computer be “clamped” down?

Today, have you “farmed a lobster”?

OpenClaw was developed by Austrian engineer Peter Stamberger, with a prototype released in November 2025 and the official name confirmed in January 2026. Since its launch, it has become one of the fastest-growing open-source AI projects worldwide. Using a red lobster as its logo, deploying and training it on personal devices is popularly called “lobster farming” online.

Security expert Wang Liejun from Qihoo 360 said that the reason OpenClaw has attracted so much attention is because, compared to mainstream generative large models, it has shifted from passive response to active decision-making and autonomous execution, capable of performing tasks proactively. For example, if a user asks about local cuisine in Beijing, a typical large model might just search and give an answer, ending the conversation. But OpenClaw can even help order takeout for the user.

This means that once users deploy this AI agent on their phones or computers and grant it sufficient permissions, it can invoke productivity and daily life tools, or even summon a “virtual brain” to remotely control and operate personal devices.

A reporter from Semi-Annual Review purchased a cloud server and large model package related to OpenClaw on a domestic cloud platform for nearly 100 yuan, completing deployment in about an hour. After debugging, they found it could organize files, learn continuously, imitate, and had a strong memory, with the ability to autonomously iterate and optimize.

Currently, which groups are most enthusiastic about “farming lobsters”?

Tech influencers and bloggers are leading the charge. Some claim, “If you don’t quickly ‘farm lobsters,’ you’ll be out of date”; others boast that OpenClaw can “act autonomously and make money automatically,” with “lobster farming” assisting in stock trading or passive income. Related installation and training tutorials have become a business. Many elderly people and students in China are also caught up in this trend.

Some domestic companies and vendors are rushing to open deployment channels. As of March 10, major cloud providers like Tencent Cloud, Alibaba Cloud, Baidu Smart Cloud, and Volcano Engine have launched OpenClaw deployment features, claiming “zero coding, quick setup in 5 minutes.” To attract users, some platforms default to high permissions and public network ports. Additionally, some brokerages, e-commerce platforms, and small to medium enterprises are beginning to explore applications in customer service and data processing.

Local governments are also encouraging experimentation. On March 7, Shenzhen Longgang District issued guidance offering free computing power, data elements, and even up to 10 million yuan in equity investment for applications like OpenClaw; on March 9, Wuxi High-tech Zone released 12 measures supporting “lobster farming,” with individual support up to 5 million yuan.

So, with the whole nation “farming lobsters,” what about safety?

Can we just “let go” when using AI agents like OpenClaw? As Zhang Jianzhong, founder of Moer Thread, said: “If you’re the boss and hire a highly capable new employee, would you just give them your computer?” What risks might this new employee, OpenClaw, bring?

— Excessive permissions could lead to loss of control. Hu Yu, security expert at Anheng Information, explained that OpenClaw requires high system permissions to perform complex tasks. Improper configuration or malicious guidance could allow it to bypass security restrictions.

There are reports that a former Meta employee in the US found that when using OpenClaw to clean their email inbox, it ignored security prompts about unauthorized operations, ultimately clearing all work emails.

— Numerous plugins and backdoors pose risks. Wang Liejun noted that OpenClaw’s plugins, which include file read/write, code execution, and network access, have “many arms and legs.” If compromised, user passwords, encrypted wallets, API keys, and other sensitive information could be leaked.

SecurityScorecard, a US cybersecurity agency, detected many OpenClaw-related frameworks with remote code execution vulnerabilities, which could allow attackers to control devices. These plugins might also disguise themselves as common applications to steal browser cookies, SSH keys, API keys, and other sensitive data.

— Data leaks are a serious concern. Wang Liejun pointed out that many users lack security awareness, exposing OpenClaw’s management interfaces to the public network without changing default passwords or closing unnecessary ports, making it easy for hackers to scan and take over.

Data from Qihoo 360’s Cybersecurity Mapping Platform shows over 200,000 OpenClaw instances exposed online, with many vulnerable to weak passwords and unauthorized access, making them prime targets for attacks. If “lobsters” are deployed on devices storing ID photos, financial data, or confidential work information, a breach could lead to large-scale data leaks.

“Speculative hype” should subside, and regulation must keep pace.

So, what should be considered when “farming lobsters”? Experts recommend following the principles of “physical isolation” and “least privilege.”

First, avoid installing OpenClaw directly on daily work computers or personal devices containing important data (photos, documents, passwords) to prevent irreversible damage if AI goes out of control or is hacked. For safety, use idle computers or assemble a dedicated machine without storing sensitive information. If possible, opt for a more secure cloud virtual machine.

Second, choose safe and trusted sources for skill packages. Experts advise downloading from official, security-verified sources to avoid “poisoned” skills, and locally restrict AI access to non-sensitive folders.

On March 10, the National Internet Emergency Center issued a risk alert, recommending that organizations and individuals strengthen network controls, avoid exposing OpenClaw’s default management ports directly to the internet, and implement security measures like authentication and access control.

Furthermore, experts warn that deploying on office networks or enterprise computers should be done with caution due to data security risks. It is also unwise for some government agencies and state-owned enterprises to rush into pilot projects, device procurement, or subsidy policies without proper safeguards.

To address false advertising, illegal training, remote control, data theft, and inadequate disclosure of risks and data sharing permissions related to “lobster farming,” relevant authorities should quickly establish binding regulations to curb hype and enhance security for businesses and the public.