OpenClaw in China: Lobster-fueled hype behind new AI agent

A remarkable spectacle resembling a bizarre festival can currently be observed across China’s major cities. Lobsters are everywhere, appearing as balloons, plush toys in claw machines or even as headwear. This is not a tribute to some obscure tradition, but rather to OpenClaw: an autonomous AI agent that is enjoying unprecedented popularity in China.

This tool enables users to hand over control of their computers and other internet-connected devices to a digital assistant that can perform tasks independently. While debates surrounding artificial intelligence in Western industrialised nations are often dominated by regulatory hurdles and ethical concerns, China is exhibiting an almost euphoric readiness to integrate this technology into daily life.

Analysts estimate that China currently has more OpenClaw users than any other country in the world. This enthusiasm is fuelled not only by the general public, but also by the state: companies receive subsidies for integrating AI assistants into their workflows, and large-scale events — sometimes drawing thousands of attendees — are frequently held to help private individuals set up OpenClaw.

A survey conducted last year highlights the difference in perception of this technology between East and West. Among Chinese workers, almost 70% of respondents believed that the benefits of using AI outweighed the risks. By contrast, only around 33% of Americans felt that the advantages of the technology outweighed the dangers. However, behind the hype lies a technical reality that can often be costly and insecure for users.

OpenClaw is based on Large Language Models (LLMs), which consume billing units known as 'tokens'. Reports indicate that even simple command loops can result in exorbitant invoices. Beyond the financial risks, there are significant cybersecurity threats: security researchers have identified a large number of unprotected OpenClaw interfaces on the internet.

Furthermore, the true level of autonomy of these systems is highly questionable. A closer technical examination of Moltbook reveals that much of this is smoke and mirrors rather than substance. The agents on this network do not act of their own accord: a human must issue the registration command and initiate every single post.