Security Frenzy Meets Secret Gambit: OpenClaws Reckoning and WeChats AI Race

The OpenClaw Frenzy and the WeChat Gambit: Security Crisis and Super-App Ambitions Reshape China's AI Agent Race

A wave of enthusiasm for "raising lobsters" has swept across China's internet, only to crash against the harsh realities of cybersecurity and intense commercial competition. At the center of this storm is OpenClaw, an open-source AI agent framework that has captured the public imagination by enabling users to create personalized, action-oriented AI assistants. However, this democratization of advanced AI has been marred by the exposure of hundreds of thousands of insecure instances, triggering official warnings and exposing a deep cognitive gap between technological fascination and practical implementation. Concurrently, in a strategic move that could redefine the landscape, Tencent is reportedly developing a secretive, powerful AI agent deeply integrated within its ubiquitous WeChat super-app, signaling a pivotal moment in the battle for the next AI "super entrance."

OpenClaw: From Niche Tool to "National Craze" and Security Headache

The phenomenon began with the release of OpenClaw, an open-source framework developed by programmer Peter Steinberger. Its breakthrough lies in transitioning AI from conversational chatbots to actionable "agents"—systems with a "brain" that can grow "hands and feet" to perform tasks autonomously based on user goals, living up to its slogan: "The AI that actually does things." Its lobster icon quickly inspired the cultural meme of "raising lobsters," referring to the process of installing, deploying, and training these AI agents with personal data.

The appeal is profound. Unlike platform-bound AI assistants from major tech firms, which operate within predefined ecosystems, OpenClaw's open-source nature allows it to interface with a vast array of local software and web tools. It promises a path toward true personalization and local/private deployment. Users can, in theory, train their "lobster" on their own data and habits, creating a digital companion that understands their preferences and executes tasks—from file management to online communications—on their behalf. This promise of a private, capable electronic partner fueled an explosive trend, with reports of queues forming for hardware and a burgeoning cottage industry of deployment services and tutorials.

However, this rapid, grassroots adoption has come with severe and immediate costs. According to data from the OpenClaw Exposure Watchboard, over 278,000 OpenClaw instances were found exposed on the public internet as of early March, with approximately 88,000 of these already suffering data breaches. The situation has grown so concerning that China's Ministry of Industry and Information Technology (MIIT) issued a public warning, highlighting the high security risks—including potential network attacks and information leaks—posed by improperly configured "lobster" instances. Some users have even reported having their accounts restricted on platforms like Xiaohongshu, Weibo, and Taobao during deployment attempts. As one cybersecurity expert grimly noted in the original report, "Twenty years ago, we called programs with such high permission requirements viruses."

Commercial Gold Rush and the "Token Black Hole"

Despite—or perhaps because of—these growing pains, the commercial potential perceived in the AI agent space has triggered a frenzied response from industry giants and investors alike. The race is framed as a battle for the ultimate "super entrance," the primary interface through which users will delegate tasks to AI.

Major Chinese tech firms have swiftly launched their own initiatives. ByteDance's Volcano Engine unveiled ArkClaw, a cloud-based SaaS version promising one-click OpenClaw deployment. Tencent introduced WorkBuddy, an AI agent designed to integrate seamlessly into QQ, Feishu, and DingTalk, and subsequently announced a full suite of "lobster"-themed products. Alibaba's Tongyi team and NetEase Youdao have also released similar open-source or agent-focused products like CoPaw and LobsterAI. Local governments, including Shenzhen's Longgang district and Wuxi, have rolled out subsidy programs to foster OpenClaw-related services, with grants reaching up to 5 million yuan.

The market excitement is driven by OpenClaw's perceived dual promise: long-term strategic positioning in the AI agent ecosystem and short-term commercial viability. A key factor is its immense computational appetite. Unlike conversational AI, an action-oriented agent like OpenClaw consumes tokens—the units of computational processing for AI models—at a rate described as orders of magnitude higher, creating a voracious "token black hole." This demand is seen as a new, lucrative commercialization avenue for AI infrastructure and model providers.

The financial markets have reacted accordingly. MiniMax, a company specializing in AI models, saw its stock price surge over 480% following its Hong Kong IPO, with its market capitalization exceeding HK$380 billion by March 10. Another standout is Moonshot AI, which achieved a valuation surpassing $10 billion after raising over $1.2 billion in two funding rounds. Its Kimi K2.5 model is a core provider within the OpenClaw ecosystem, and its API revenue reportedly eclipsed its total 2025 income in just 20 days post-launch.

The Unmasked Vulnerability: A Crisis of Misconfiguration

The security crisis surrounding OpenClaw is not a flaw in its core code, but a widespread failure in its deployment and configuration—a direct consequence of lowering the barrier to entry without a parallel rise in security literacy. The hundreds of thousands of "naked" instances represent deployments where the OpenClaw agent, often requiring high system permissions to perform its designated tasks, has been left exposed to the public internet without adequate authentication, firewalls, or access controls.

This makes them low-hanging fruit for malicious actors. Compromised instances can lead to severe consequences: theft of the private data used to train the agent, unauthorized access to connected user accounts and services (like email or social media), co-option of the agent to perform malicious actions on the user's behalf, or the hijacking of computational resources. The MIIT warning underscores that this is not a theoretical risk but an active threat, with data breaches already confirmed.

The situation reveals a critical "cognitive裂缝" (cognitive gap): users are captivated by the functional promise of AI agents but lack awareness of the operational responsibilities they entail. The drive to participate in the trend has outpaced the understanding of running a complex, internet-connected software service securely. This gap presents a significant challenge for the sustainable development of the democratized AI agent movement, potentially eroding trust and inviting stricter regulatory scrutiny.

WeChat's Cautious Power Play: Embedding AI in the "Super Operating System"

As the OpenClaw community grapples with security, the established platform giants are plotting their next moves. According to a report by The Information, Tencent is developing a "top-secret" AI agent project designed to be deeply and natively embedded within WeChat. This initiative, reportedly酝酿 since the first half of 2025, aims to transform WeChat from a messaging and mini-program platform into a proactive "life manager."

The project's core strategy leverages WeChat's unparalleled ecosystem. Instead of being a standalone app, the AI agent would operate within a chat interface, allowing its 1.4 billion monthly active users to delegate complex, multi-step tasks involving the millions of services available through WeChat Mini Programs. The envisioned scenario involves a user asking, "Take my family to Hangzhou for the weekend and book high-speed rail tickets and a hotel near West Lake," with the AI autonomously handling search, comparison, booking, form-filling, and payment across different mini-program services like Ctrip or Meituan.

This approach tackles the "super entrance" battle from a position of unique strength. While Alibaba's Tongyi and ByteDance's Doubao have deeply integrated their AI with their own e-commerce and service ecosystems, they remain applications users must actively open. WeChat is already the default "operating system" for daily digital life in China. Integrating a powerful agent directly into this flow removes the need to acquire users for a new product; they are already present and engaged.

However, the report also highlights strategic challenges that have slowed WeChat's AI progress. An overriding concern for Tencent has been the risk of rolling out immature features to its massive, entrenched user base, where any misstep could cause widespread negative feedback. Furthermore, technical compromises are being made: the team is reportedly testing models from external providers like Zhipu AI, Alibaba, and DeepSeek, as well as its own smaller models, rather than relying solely on Tencent's in-house Hunyuan model, whose performance is perceived as not yet leading the industry. This reliance on third-party models may complicate data integration and authorization processes, affecting development timelines. The project is currently slated for a limited beta around mid-year, with a full launch possible in the third quarter, though WeChat's "not stable enough, not released" philosophy could cause delays.

Converging Paths, Divergent Strategies

The concurrent narratives of the OpenClaw frenzy and WeChat's secret project illuminate two powerful, competing vectors shaping the future of AI agents. OpenClaw represents a bottom-up, decentralized, and user-empowering vision. It unlocks creativity and personalization but is currently plagued by the security and sustainability challenges inherent in any rapidly democratized powerful technology.

WeChat's approach represents the top-down, platform-centric model. It seeks to leverage an existing monopoly over user attention and digital services to deliver a seamless, integrated, and potentially more stable assistant experience, albeit within the boundaries of its own walled garden.

The massive commercial investments and regulatory attention now focused on this space indicate that AI agents are moving swiftly from conceptual hype to a fraught but decisive phase of implementation. The race is no longer just about who has the best model for conversation, but who can most effectively, securely, and reliably bridge the gap between AI instruction and real-world action. The outcome will depend not only on technological prowess but on managing the critical trifecta of security, usability, and ecosystem scale. As the "lobsters" are secured and WeChat's agent prepares for its debut, the battle for the AI super entrance is entering a new, more concrete, and complex chapter.