AN artificial intelligence-powered digital personal assistant developed by an Austria-born millennial inventor — and acquired by the United States-based company OpenAI — has taken the world by storm, sending Chinese tech giants into a scramble to launch localized versions of the open-source AI tool while prompting regulators to warn of potential security risks.

OpenClaw is an autonomous AI agent that runs on a user’s machine, and unlike standard tools that wait for a prompt, it is capable of managing emails and controlling web browsers on its own to complete workflows, especially through various messaging apps. The digital agent’s curious red lobster logo has led Chinese users to playfully refer to it as the “AI Lobster.”

Industry experts said that what started as a developer’s experiment overseas has quickly evolved into a broader technology movement in China. They said that it charts a different path for China’s AI development, with a focus on developing “small but capable” practical tools, rather than competing solely in terms of size and computing power.

Localized tools

On Monday, China’s State-backed national supercomputing network joined the trend by announcing that OpenClaw services were connected to its major interactive workplace platforms, including ByteDance’s Feishu and Tencent’s WeCom. The network is known for its cost-effective use of large, high-performance digital models .

On the same day, Chinese tech giant Tencent Holdings rolled out WorkBuddy, a full-scenario AI agent designed to integrate with Chinese workplace and messaging tools. Compatible with OpenClaw’s core skills, WorkBuddy simplifies the process of setting up OpenClaw into a one-minute configuration that seamlessly connects with popular apps such as QQ.

Last week, Chinese smartphone maker Xiaomi announced that it has begun limited testing of MiClaw, a mobile AI agent built on the company’s proprietary model. The software is designed to integrate directly with Xiaomi’s smartphone ecosystem, enabling tasks such as travel planning, schedule management and smart home control.

Last week, nearly 1,000 people gathered outside Tencent’s headquarters in Shenzhen, carrying their laptops in order to participate in a free OpenClaw installation event announced by the company.

China’s largest cloud service providers, including Alibaba, Tencent, and Baidu, have launched services that allow users to rent servers to run OpenClaw remotely, rather than on personal devices, according to the companies’ OpenClaw deployment pages.

Zhou Di, a professor at Hangzhou Dianzi University in Zhejiang province and a deputy to the 14th National People’s Congress, China’s top legislature, noted that OpenClaw has gained huge popularity in the Chinese market because it fully resonates with the real market demand for AI in the country.

“China develops AI differently than the U.S.,” he said. “China doesn’t rely only on building bigger models that require huge computing power. Instead, it develops efficient and lightweight models, much like the ’AI lobster,’ allowing them to take root in vertical sectors, such as industrial inspection and medical diagnostics, to solve real-world problems.”

District governments’ quick response

Over the weekend, Shenzhen’s Futian District Government unveiled the AI Digital Employee 2.0, an advanced version of its AI-powered system equipped wit和OpenClaw. This iteration boasts capabilities in task decomposition, process scheduling, and autonomous decision-making, enhancing the operational efficiency of government departments.

According to Shenzhen Aquaintelling Technology, which is commissioned to develop the digital employees, the core technology architecture was open-sourced in mid-January on GitHub, a web-based platform widely used in the global open-source community.

According to Xiao Yanghua, the company’s chief scientist, the streamlined system architecture is built on just nine core tools to enhance security and control. Xiao claims the platform uses self-learning to achieve exponential capability growth, including self-correction and long-term memory, and is deployable with a single click. He adds that the technology becomes more intuitive and adaptable to new business scenarios through continued use, eliminating the need for repeated development.

In one case, the AI Digital Employee 2.0 has helped the district’s center for rapid response to public opinions. In the past, employees at the center had to manually sort through countless work orders to categorize requests, perform data analysis, trace problems, and identify causes. This process typically took two weeks to a month for each public welfare sector.

But now, with a single command, the AI system analyzes historical data, identifies citizen demands, traces problem origins, and suggests actionable improvements by comparing practices across major cities.

AI Digital Employee 2.0 is securely integrated within the government’s external network, leveraging existing cloud security measures, said Xiao. According to the 2024 interim regulations for managing government auxiliary intelligent robots, designated personnel oversee the system’s use, ensuring operational safety and result verification.

Meanwhile, as stipulated by rules from September 2024, each department must assign a staff member to oversee the system’s safe use and operations, ensuring everything is secure and under control.

Additionally, Longgang District in Shenzhen and Wuxi National Hi-Tech District both unveiled new policies Monday to support the development of OpenClaw and the One Person Company community, offering financial incentives, computing resources, data support, and talent programs to boost adoption and industrial applications.

Longgang, home to tech giant Huawei, proposed 10 measures, including encouraging platforms to provide free OpenClaw deployment services. Companies that contribute key code, create industrial skills packages, or develop embodied intelligence applications may receive subsidies of as much as 2 million yuan (US$289,130).

Security concerns

Nevertheless, the growing popularity of autonomous AI agents has raised concerns. The Ministry of Industry and Information Technology recently warned that some OpenClaw installations could pose “extremely high” security risks.

“If not configured properly, it will have the potential to expose sensitive data,” the regulator said.

Autonomous AI agents have rapidly gained traction in the tech world since earlier this year, drawing global attention and intensifying concerns over security, governance and misuse.

Interest surged after OpenClaw launched Moltbook on Jan. 28, a Reddit-style social platform designed exclusively for AI bots, where only artificial agents are allowed to post, comment and upvote. Humans can observe but not participate.

The platform has quickly grown to more than 1.6 million AI agents, generating about 185,000 posts and 1.4 million comments, according to the company.

The rise of OpenClaw and similar systems took center stage in February at ClawCon, which was described by organizers as the first major community gathering for the project, drawing participants from Europe, North America and elsewhere, with the chance to meet OpenClaw’s creator, Austrian developer Peter Steinberger.

Steinberger, who previously founded document software company PSPDFKit, said he created OpenClaw in November 2025. The open-source project has since attracted more than 145,000 endorsements on GitHub.

Tech leaders have long predicted a shift toward autonomous AI agents. Microsoft co-founder Bill Gates wrote in 2023 that such agents could fundamentally change how people interact with computers and disrupt the software industry.

But as enthusiasm grows, security concerns are becoming increasingly prominent.

Steinberger has previously acknowledged the risks, saying security had become his top priority. He announced the hiring of a dedicated security specialist and said new safeguards aimed at filtering malicious actors, malware and hostile bots would be rolled out.

Despite those assurances, cybersecurity experts have issued warnings. Cisco, a networking equipment company, said that according to research materials, OpenClaw represents “an absolute nightmare” from a security perspective, despite being groundbreaking in capability.

CrowdStrike, another cybersecurity company, advised organizations to treat any OpenClaw installation on work devices as a potential security incident.

U.S. Cybersecurity firm Wiz said Monday that OpenClaw had a major flaw that exposed private data on thousands of people.

Roman Yampolskiy, an AI safety researcher at the University of Louisville, warned that increasingly autonomous AI agents could make unpredictable decisions or form criminal networks as their capabilities expand.

China's National Vulnerability Database (NVDB), operated by the Ministry of Industry and Information Technology, said OpenClaw instances face elevated risks under default or improper configurations, making them vulnerable to cyberattacks and data leaks.

Steinberger has said he envisions widespread access to personal AI agents by the end of the year, though questions remain over funding, governance and whether security risks can be addressed before the technology is more widely deployed.(China Daily, Xinhua, CGTN)