Overview
In early 2026, an AI craze called “lobster farming” spread rapidly across the Chinese internet, drawing attention from tech giants to ordinary netizens. The farming is metaphorical: it refers to deploying and tending OpenClaw, an open-source AI agent. The free tool, built by Austrian developer Peter Steinberger and named for its red-lobster logo — Chinese users affectionately call it “the lobster” — turned deployment into a hobby, complete with the slang of animal husbandry. From tech giants like Tencent and Baidu joining in, to Shenzhen’s Longgang district launching the “Ten Lobster Measures” worth millions of RMB, to China’s authorities issuing emergency risk warnings, OpenClaw’s viral moment in China was more than a technology story. It refracted the arrival of the AI agent era — with its opportunities and hazards in full view.
What Is OpenClaw, and How Is It Different From Traditional AI?
OpenClaw is free, open-source AI agent software first published on GitHub in November 2025. Unlike a conventional large language model, it is not merely conversational — it is an automated assistant that operates the computer directly. Its “heartbeat mechanism” lets it proactively execute tasks on a schedule the user defines: checking email, reading news, managing calendars, sending messages, even driving the browser and the file system.
The distinction in one line: traditional LLMs like ChatGPT and Gemini answer questions; OpenClaw receives a task, plans its own steps, calls external tools — browsers, APIs, databases, search engines — and actually executes. Users interact with it through Telegram, WhatsApp, and similar messengers, delegating everything from organizing files and drafting articles to booking flights and managing itineraries.
One caveat matters: OpenClaw itself is free, but it runs on commercial AI models billed per token. Every action — reading the news, deleting a file — carries a cost. Some users set a $20 daily spending cap per assistant; lobster farming, it turns out, needs real feed money.
Why Did “Lobster Farming” Explode in China?
The boom had several engines. First, technological anxiety and FOMO. Against a backdrop of employment pressure and rapid AI iteration, many Chinese netizens joined the craze for fear of being left behind. Xiaohongshu, Douyin, and Weibo filled with tutorials and install guides, and some developers sold door-to-door deployment services — a complete “install-for-you” industry chain.
Second, China’s tech giants poured fuel on the fire. Baidu hosted a “lobster market” at its Beijing headquarters, where engineers installed OpenClaw on the cloud for free; Tencent’s Shenzhen install event drew queues of nearly a thousand people, after which Tencent Cloud announced free deployment, model configuration, and skill installation across 17 cities. Tencent also launched SkillHub, an AI-skills community optimized for Chinese users, and became an official OpenClaw sponsor; Baidu shipped the first lobster phone app.
The most remarkable push came from policy. In March 2026, Shenzhen’s Longgang district issued the “Ten Lobster Measures,” aiming to make the district the world’s first choice for AI agent startups: development grants up to 2 million RMB for OpenClaw projects, up to 4 million RMB in compute and application-scenario support, equity-investment channels up to 10 million RMB, plus talent-settling subsidies, free office space, and startup rent discounts. With that, lobster farming ascended from a grassroots craze to an officially backed industrial direction.
The Security Turn: From Farming to Uninstalling
Beneath the boom, undercurrents moved. As users multiplied, OpenClaw’s security risks surfaced. China’s national computer emergency response center warned of serious incidents caused by improper installation and use: stolen photos, documents, and chat logs; leaked API keys and payment accounts; and agents misreading instructions and deleting mail and data outright.
A body under the Ministry of Industry and Information Technology issued its own alert, citing blurred trust boundaries, privilege escalation beyond control, sensitive-data leakage, and an enlarged external attack surface. Bloomberg reported that Chinese state-owned enterprises and government agencies received notices restricting OpenClaw on office networks, and at least 15 Chinese brokerages issued internal compliance notices banning unauthorized installation on office or business systems.
The specific threats: OpenClaw typically needs elevated system permissions, so attackers can plant prompt-injection payloads in web pages to coax the agent into reading sensitive data; a serious vulnerability, CVE-2026-33579, let a least-privilege user promote themselves to administrator; and attackers distributed info-stealing malware through fake GitHub repositories.
The risks drove a dramatic pivot from “farming the lobster” to “uninstalling the lobster.” Many users, disappointed by the technical threshold, the running costs, and the modest real-world utility, scrambled for next-day removal services — and a cottage industry of paid “de-lobstering” merchants appeared. As one Hong Kong tech-association chairman put it bluntly: OpenClaw is “not very practical, expensive to run, with risk out of proportion — its greatest value is that it’s fun.”
Outlook: Where AI Agents Go Next
The OpenClaw craze cooled under the double weight of security controversy and underwhelming user experience. But it indisputably opened the agent era’s first chapter. Founder Peter Steinberger announced he would join OpenAI for research while handing OpenClaw to a foundation for maintenance. Industry consensus expects more enterprise-customized “lobster” versions that lower the technical bar and raise the security floor.
Hong Kong DotAI’s chief executive noted that tuned AI assistants keep improving and can follow a company’s internal operations indefinitely — no resignations, no raise demands — a structural advantage over human staff. And OpenClaw’s open-source nature means any enterprise can build its own application version on the codebase, forming canned, standardized services that will likely be the real path to agent adoption.
The trendline underneath the hype is unambiguous. Set aside the giants’ active positioning: Shenzhen’s policy support and global attention to agent technology all point the same way: whatever happened to lobster farming’s popularity curve, the paradigm shift agents represent — from answering questions to executing tasks — is not reversible. As BBC Chinese summarized it: the craze will mark the future direction of AI.
The Bottom Line
China’s lobster-farming boom was a multi-dimensional phenomenon braiding technical innovation, social anxiety, commercial positioning, and policy gaming. It showed AI agents’ potential to cross from concept into practice — and reminded everyone that embracing new technology requires taking security and risk control seriously. For ordinary users, the decision to “farm a lobster” deserves an honest audit of technical threshold, running cost, and security exposure. For enterprises and developers, the agent wave has already arrived, and whoever first balances safety, usability, and cost will hold the initiative in the next round of AI change. The craze may fade. The agent future is just beginning.
Tags: AI agent, OpenClaw, lobster farming, artificial intelligence, open-source software, cybersecurity, Tencent, Baidu, technology trends
