Dawn of the AI Agent Era, Global Momentum Persists Despite China’s ‘OpenClaw Security Controversy’

OpenClaw Craze Sweeps China, AI Agents Adopted Even by Public Institutions
“Security Threats Exist” — Popularity Cools Rapidly Amid Government Warnings
Robust Global Demand for AI Agents, Big Tech Including Meta Accelerates Development

The artificial intelligence (AI) agent market is expanding at a rapid pace. As AI utilization capabilities emerge as a core competency in white-collar workplaces, interest is increasingly shifting toward AI agents, which offer superior task execution capabilities compared to conventional chatbots. In China, ‘OpenClaw’ gained explosive popularity, followed by a wave of similar services entering the market, while major big tech firms such as Meta are accelerating the development of proprietary AI agents to enhance operational efficiency.

OpenClaw Sweeps the Chinese Market

According to the IT industry on the 24th, an ‘OpenClaw craze’ has swept across China in recent weeks. OpenClaw is an open-source AI agent developed by Austrian engineer Peter Steinberger, capable of autonomously performing tasks such as file management, web browsing, and code execution on a user’s computer. It effectively functions as an ‘AI assistant’ that goes beyond simple conversational chatbots by holding partial OS-level permissions and executing real tasks.

The impact of OpenClaw was particularly pronounced in Shenzhen, home to China’s major big tech firms. On the 6th, more than 1,000 people gathered in front of the Tencent Building in Shenzhen for a free OpenClaw installation event. Several local government offices in Shenzhen also adopted OpenClaw for practical use and reported tangible outcomes. Futian District, for instance, utilized OpenClaw to rapidly analyze large volumes of civil complaints, while Longgang District announced proactive support policies aimed at fostering OpenClaw development.

Chinese big tech companies also responded to the OpenClaw phenomenon by launching their own AI models and related services. Tencent introduced ‘Workbuddy,’ an AI assistant integrated with office and communication tools, while ByteDance, the parent company of TikTok, unveiled ‘ArkClaw,’ a cloud-based AI tool requiring no separate installation. Alibaba developed ‘CoPaw,’ an AI agent integrated with enterprise messaging platforms such as DingTalk and Feishu, while Chinese AI startup Zhipu released ‘AutoClaw,’ software designed to simplify the installation process of OpenClaw.

Security Risks Surface Amid Government Restrictions

However, OpenClaw’s popularity has recently begun to decline sharply. Security controversies surrounding the platform have dampened market sentiment. On the 11th (local time), Bloomberg, citing multiple sources, reported that government agencies and state-owned enterprises, including China’s largest banks, had issued warnings against installing OpenClaw software on office computers due to security concerns. Some officials were reportedly instructed to undergo security inspections if they had already installed the application and to delete it if necessary.

Individual users are also increasingly refraining from using OpenClaw. Paid services offering ‘complete deletion’ of OpenClaw have even emerged on Chinese social media and e-commerce platforms. This trend is interpreted as a response to growing concerns expressed by local authorities over the proliferation of AI agents. The National Vulnerability Database (NVD) under the Ministry of Industry and Information Technology (MIIT) issued usage guidelines for OpenClaw this month, recommending the use of the latest official version, limiting internet exposure, and restricting permission settings. Another MIIT-affiliated institution, the China Academy of Information and Communications Technology (CAICT), is establishing standards to address the opaque decision-making processes of OpenClaw agents and plans to introduce ‘Intelligent Assistant Agent Product Reliability Requirements’ to ensure quality control and behavioral reliability of AI agents.

China’s cautious stance toward OpenClaw reflects the fact that its risks have already been substantiated. Last month, cybersecurity firm DepthFirst disclosed via its official blog that it had discovered a one-click remote code execution (RCE) vulnerability in OpenClaw. This represents a critical security flaw that allows attackers to fully compromise a computer with a single click on a malicious webpage. In the same month, Bloomberg reported a case involving Chris Boyd, a software engineer based in North Carolina, who encountered issues after granting OpenClaw access to iMessage. Boyd stated that after granting permission, OpenClaw suddenly became uncontrollable, sending over 500 messages to him and his wife and dispatching spam messages to random contacts.

The ‘Potential’ of AI Agents

Although the OpenClaw frenzy in China has effectively ended as a short-lived episode, market consensus suggests that global demand for AI agents will continue to strengthen. IT research and consulting firm Gartner projected in its latest report that by this year, approximately 40% of enterprise software could incorporate AI agents. This outlook is driven by the rise of ‘AI delegation’—the ability to assign tasks to AI and evaluate outcomes—as a critical competency for white-collar workers. Gartner further forecasts that by 2028, AI will autonomously handle 15% of routine work decisions.

Development and deployment of AI agents by major big tech firms are also accelerating. Meta, for example, is building an AI agent designed to support CEO Mark Zuckerberg’s managerial tasks. According to multiple media reports, the agent aggregates vast amounts of internal and external data in real time, synthesizing insights necessary for executive decision-making and enabling direct access to critical information without intermediary reporting processes. Some analysts suggest that Meta’s CEO agent could function as a central hub integrating internal personal task agents such as MyClaw and project documentation indexing tools like Second Brain.

Meta’s focus on such initiatives stems from demonstrable productivity gains associated with AI adoption. In January, Axios reported that following the introduction of internal AI coding tools, Meta engineers’ productivity increased by an average of 30%, with top power users achieving up to 80% improvements. A Meta engineering leader also confirmed these figures in a LinkedIn analysis, emphasizing that “these are not estimates but measured data derived from Meta’s workforce of approximately 79,000 employees.”

Having confirmed the tangible potential of AI, Meta is not only strengthening its technological capabilities but also significantly expanding its investment scale. The company’s capital expenditure (CAPEX) guidance for this year is projected at between $115 billion and $135 billion. This represents nearly a doubling compared to last year’s guidance of $72 billion. The majority of the additional investment is expected to be allocated to AI-dedicated infrastructure.