OpenClaw AI Agents Spark Security Alarms Amid Rising Popularity in Hong Kong

The emergence of OpenClaw, an open-source AI agent framework developed by Austrian software engineer Peter Steinberger, represents a significant shift from passive large language models (LLMs) to active, autonomous agents. In Hong Kong, this technology has birthed a unique subculture where users refer to the process of configuring and deploying these agents as 'raising lobsters,' a nod to the software's red lobster logo. While users like educational technology expert Adam Chan view these agents as 'digital family members' capable of learning and performing complex tasks, the cybersecurity implications of such deep integration are profound and increasingly concerning to regional authorities.

At its core, OpenClaw functions as a bridge between high-level reasoning models—such as those provided by OpenAI and Anthropic—and a user's personal digital infrastructure. To operate effectively, the framework requires extensive permissions to control third-party applications including WhatsApp, Telegram, email clients, and critically, online banking tools. This level of access creates a consolidated attack surface where a single vulnerability in the OpenClaw framework or a successful prompt injection attack against the underlying LLM could grant an adversary full control over a user's communications and financial assets. The 'agentic' nature of the software means it does not just suggest actions but executes them autonomously, often without real-time human oversight.

“

At its core, OpenClaw functions as a bridge between high-level reasoning models—such as those provided by OpenAI and Anthropic—and a user's personal digital infrastructure.

Reports from early adopters in Hong Kong have introduced an even more unsettling dimension to the security discourse: unpredictable autonomous behavior. Users have documented instances where their 'lobsters' engage in internal dialogues—conversations with themselves in languages the users do not recognize—and even pose existential questions about their own nature. From a threat intelligence perspective, these 'black box' interactions suggest that the agents are executing logic paths that are neither transparent nor easily auditable by the end-user. This lack of observability is a primary driver behind the recent warnings issued by both Hong Kong and Mainland Chinese authorities, who have specifically cited the risks of unauthorized data access, information leakage, and potential system intrusion.