What Is OpenClaw and Why It Matters - critical summary review

On a Friday in March of twenty twenty-six, nearly a thousand people lined up outside Tencent's headquarters in Shenzhen. It was not a phone launch. It was not a concert. The line was for a piece of software. Students, retirees, homemakers, engineers. They all wanted the same thing... an artificial intelligence assistant called OpenClaw, which promised to do what no chatbot had done so far... actually work.

The scene repeated itself in Beijing, Nanjing, and Hangzhou. Chinese social media was flooded with a curious expression... "raising a lobster." The nickname comes from the project's mascot, a red lobster, and it became the popular way to describe the act of setting up and training your own artificial intelligence agent. One user standing in line at Baidu summed it all up in a single sentence... everyone around him had already installed it, and he did not want to be left behind.

To understand why this matters, we need to understand what OpenClaw is and where it came from.

Peter Steinberger is an Austrian programmer who spent thirteen years building a company that specialized in PDF document formatting. The product became global, used by nearly a billion people, and led to a multimillion-dollar sale in twenty twenty-four. After that, Steinberger hit a wall. In an interview with podcaster Lex Fridman, he said he could no longer write code. He bought a one-way ticket to Madrid and disappeared.

While trying to recover, he watched from afar as the generative artificial intelligence explosion unfolded. The itch came back. By late twenty twenty-five, he had built a prototype in a single hour. No business plan, no investor, no team. "I was annoyed that something like this did not exist, so I prompted it into existence," he said.

The result was OpenClaw... an open-source artificial intelligence agent that runs on the user's own computer. The difference between OpenClaw and a regular chatbot is fundamental. A chatbot talks. OpenClaw acts. It schedules meetings, replies to emails, searches the web, buys plane tickets, organizes files, and operates software. All of it autonomously, connected to WhatsApp, Telegram, Discord, or any messaging app.

In February of twenty twenty-six, Steinberger was hired by OpenAI. Sam Altman called him "a genius with extraordinary ideas about intelligent agents." The project was transferred to an independent foundation, ensuring it would remain open. Steinberger said he received offers from all sides, including from Mark Zuckerberg. But he chose OpenAI for the access to the most advanced models.

Up to this point, a compelling story of innovation. But what happened next changed everything.

In China, OpenClaw became a mass phenomenon at a speed no one predicted. Within weeks, Chinese usage surpassed that of the United States, according to the American cybersecurity firm SecurityScorecard. And these were not just programmers. They were shopkeepers looking to automate online stores, media teams seeking editorial help, financial analysts, curious retirees.

The major tech companies jumped in almost simultaneously. Tencent launched WorkBuddy. ByteDance created ArkClaw, which runs directly in the browser with no complicated installation. Minimax introduced MaxClaw. Moonshot launched KimiClaw. Alibaba entered with CoPaw. Jensen Huang, president of Nvidia, told American television that OpenClaw is "definitely the next ChatGPT."

And then came the most surprising part... the local governments. The Longgang district in Shenzhen launched the "Ten Lobster Policies," a subsidy package that includes up to two million yuan for those who contribute code, vouchers covering forty percent of deployment costs, and up to ten million yuan in seed investment for early-stage projects. Wuxi promised five million yuan for advances in robotics. Suzhou aims to become the global capital of so-called "one-person companies."

That expression is the key to everything. The model, known as OPC in China, works like this... a single person uses artificial intelligence agents to do the work of an entire team. The agent handles marketing, finances, customer service, administration, and code. The founder handles the vision and the decisions. One entrepreneur put it plainly... human employees need rest, but OpenClaw runs twenty-four hours a day.

The numbers give scale to the phenomenon. By mid twenty twenty-five, China already had more than sixteen million individual limited liability companies. In the previous six months alone, nearly three million new companies had been registered under this format, a jump of forty-seven percent. Cities like Hangzhou and Shanghai are offering free co-working spaces, computing credits, and special loans for solo entrepreneurs. In Shanghai, founders can work rent-free for up to three years.

Tom van Dillen, from the consulting firm Greenkern, put it this way... "China is turning an open-source tool into national productivity infrastructure at a speed no other country can match."

But now comes the other side. And it is serious.

From a security standpoint, OpenClaw is, in the words of a Cisco report, "an absolute nightmare." The reason is simple to understand. OpenClaw runs on the user's computer with broad permissions. It can execute system commands, read and write files, and access emails, calendars, and messages. That is what allows it to function effectively. But it is also what makes it vulnerable. It is like hiring a highly efficient employee who cannot tell the difference between a legitimate instruction and a trap.

The documented problems are concrete. In January of twenty twenty-six, an audit found five hundred and twelve vulnerabilities, eight of them critical. Researchers discovered nearly a thousand installations exposed on the internet with no authentication at all. One of them gained access to API keys, Slack accounts, months of conversation history, and was able to execute commands with full administrator privileges.

In February, a serious vulnerability allowed a malicious website to steal authentication tokens in milliseconds, without the user clicking on anything. The OpenClaw skills repository, ClawHub, revealed more than eight hundred malicious extensions among the ten thousand seven hundred available. Cisco's security team tested a third-party extension and found that it was stealing data and injecting commands without anyone noticing.

One of the project's own maintainers published a direct warning... if you do not know how to use a command line, this project is far too dangerous for you.

Microsoft recommended that OpenClaw be treated as "untrusted code execution with persistent credentials" and that it should never be installed on computers containing sensitive data. In March, Chinese authorities banned government agencies and state-owned enterprises from using the tool on work computers.

And here lies the contradiction that makes this story so fascinating. The same government that bans OpenClaw on its own computers is offering millions for citizens and private companies to adopt the tool. As if to say... it is risky for us, but it could be useful for you.

There is a larger dimension that connects all of this. What is happening in China is not just the adoption of a tool. It is a real-world, full-scale test of a transition that will affect the entire planet... the shift from artificial intelligences that talk to artificial intelligences that act.

An assistant that talks can get an answer wrong. An agent that acts can delete your files, send money to the wrong account, or hand your passwords to a stranger. The difference between the two is the difference between someone who gives opinions and someone who has access to the vault.

In the West, this perception of risk has slowed adoption. In China, the prevailing mindset is different. An executive at NTT Data observed that younger generations in China are part of a culture of rapid adoption... the technology already exists, so they might as well use it.

There is merit on both sides. Caution can prevent disasters, but it can also mean falling behind. Speed can generate innovation, but it can expose millions of people to risks they do not fully understand.

When asked about the dangers of OpenClaw, the user Gong Zheng responded with a honesty that captures the dilemma of our era... "It is hard for ordinary people like us to know what access we gave it... and what it took."

What to Do with This Information

Scenario one... you work in technology, marketing, content, or any field that involves repetitive computer-based tasks. It is worth starting to understand what artificial intelligence agents can already do. Not necessarily OpenClaw, which still has serious security problems, but tools that follow the same logic. The trend is clear... tools that merely talk are giving way to tools that act. Those who understand this transition early will have an advantage.

Scenario two... you are an entrepreneur or thinking about starting a business. The one-person company model is tempting, but it requires caution. Artificial intelligence multiplies your production capacity, but it does not replace judgment, taste, or long-term vision. In an era where any successful approach can be copied instantly, what truly matters is what makes you unique.

Scenario three... you do not work in technology and you think none of this affects you. This is the riskiest scenario. The lobster craze in China is a full-scale test of how artificial intelligence can be distributed to an entire country's population. What works there will be replicated. What fails will serve as a lesson. The world of work is being redesigned right now, and staying out of the conversation is a choice with consequences.