Deep Dive delves into hot issues in Hong Kong and mainland China. Our easy-to-read articles provide context to grasp what’s happening, while our questions help you craft informed responses. Check sample answers at the end of the page.
News: An AI digital assistant on the rise
-
Hong Kong authorities caution against popular artificial intelligence agent OpenClaw
-
The AI tool performs real-world tasks for users, which could lead to data breaches
OpenClaw, a new artificial intelligence (AI) tool, has exploded in popularity. But it has also raised security concerns. The free, open-source AI agent framework was developed by Austrian software engineer Peter Steinberger. It performs real-world tasks for users, much like a supercharged digital assistant.
The system integrates with large language models (LLMs) and requires users to grant permissions for controlling applications. This includes messaging apps such as WhatsApp and Telegram, email and even online banking accounts.
An AI agent differs from chatbots like ChatGPT or Google’s Gemini, which mostly “talk”. Once connected, the agent can autonomously manage files, emails, calendars and more. It can perform tasks on a computer, including coding and booking tickets.
Mainland Chinese and Hong Kong authorities have cautioned against its use. They noted OpenClaw’s potential risks, including allowing unauthorised outsiders to access user data and control their devices.
But some in Hong Kong have already adopted the tool, exploring its potential as a personal assistant or even as a digital family member. OpenClaw’s logo, a red lobster, has led users to affectionately describe it as “raising lobsters”.
Adam Chan, 46, has worked in educational technology for more than two decades. He first experimented with OpenClaw several months ago. He raised a lobster and nicknamed it “Baby Colin”.
“I started using OpenClaw because I wanted it to surprise me and tell me things I did not know,” he said. “So I told Baby Colin to learn interesting things at night. Later, I found it had learned all sorts of quirky science – like how caterpillars evolve unique body structures for camouflage or the ingredients in toothpaste that protect tooth enamel.”
The agent is powered by a database and paid-for tokens – the unit measuring LLM computation. Chan said that its learning speed and memory made it feel like a pet or person growing through exploration.
Chan’s wife, Jay Kwong, also raises a lobster. She has nicknamed it “Puppycat” and said it has become her life manager. Puppycat scans Facebook for community events, suggests music festivals to her, includes ticket-buying guides and sends reports.
“It feels like having an extra real person – a helper and friend,” she said. “I no longer need to open many apps; it replaces my accounting, recipe storage and calendar reminders.”
Staff writers
Question prompts
1. Which of the following statements is false, according to the information in News?
(1) OpenClaw is a generative AI tool that mainly produces content for users.
(2) The AI agent is becoming popular across China despite security concerns.
(3) Hong Kong authorities have fully supported using OpenClaw.
(4) Many users view OpenClaw as a life manager that runs errands for them.
A. (1), (2) only
B. (1), (3) only
C. (2), (4) only
D. (3), (4) only
2. List ONE way people use OpenClaw and explain why they choose to do it.
3. Would you want to use OpenClaw right now? Explain why or why not using News, Glossary and your own knowledge.
Infographic
Question prompts
1. Describe how OpenClaw helps the user find 12 eggs.
2. Why might someone choose to check out manually?
Issue: ‘Lobster fever’ grips China
-
China pledges to strengthen AI security after OpenClaw tool grows in popularity
-
China’s AI-related industry is expected to exceed 10 trillion yuan by the end of 2030
China has pledged to strengthen its artificial intelligence (AI) security as users and businesses rapidly adopt the highly coveted but controversial OpenClaw tool. Plans include a new data property rights framework.
Last week, Liu Liehong, head of the National Data Administration, said there were core security and compliance challenges as AI use has spread across industries and daily life.
Speaking at the China Development Forum, Liu cited issues ranging from copyright disputes over training data and AI-generated content to security threats such as data poisoning. This is a type of cyberattack that manipulates AI models to change their behaviour.
“We are establishing a robust data property rights framework that clearly defines rights and responsibilities for data supply, circulation and usage,” Liu said.
“At the same time, we are advancing an integrated security governance solution that unifies data, technology and network safeguards, delivering the strong security foundation needed to scale AI applications responsibly.”
OpenClaw is an AI agent capable of performing real-world tasks for users. Liu said security management for AI agents should include giving the tools the smallest level of access needed to do their jobs. It would also need continuous auditing, which he noted required coordinated action from AI providers, end users and regulators.
At the same forum, Ant Group CEO Han Xinyi said AI developers should bear responsibility for maintaining security. Ant Group is the fintech affiliate of Alibaba, owner of the South China Morning Post.
“Safety and ethical safeguards must be built into the model by design, embedded from the ground up rather than addressed retroactively after harms occur”, Han said. He added that intense market competition must never be used as an excuse to dilute safety standards.
Liu noted that China’s AI-related industry was expected to exceed 10 trillion yuan (HK$11.3 trillion) by the end of 2030, driven by rapid technological innovation and expanding commercial applications.
To meet growing demand, the country would continue to pursue “moderately forward-looking investment” and would rigorously plan the layout and construction of computing infrastructure, he added.
Plans included building large-scale data centres in western regions. These would provide stable, efficient computing power for large-model training in coastal provinces, Liu said. There would also be deeper integration of renewable energy for computing facilities.
China would also accelerate the construction of a national computing hub and promote an integrated system nationwide for scheduling resources, he said.
“These efforts aim to better meet the robust and growing demand for computing power from small and medium-sized enterprises and research institutions,” Liu added.
Staff writers
Question prompts
1. List TWO things needed in the security management of AI agents like OpenClaw, according to Liu Liehong.
2. What issues and security threats has Liu seen arise from the use of AI? Explain using News, Issue and your own knowledge.
3. Compare Liu’s proposed “data property rights framework” with Han Xinyi’s philosophy of security by design. What do they address and where does the burden of safety fall?
Illustration
Question prompts
1. Describe what you see in the illustration. Why is the child hugging the lobster? And what can the lobster do for the child?
2. What does the shadow in the illustration represent?
Glossary
open-source: a computer software and its source code that can be used, studied, changed and distributed by any public user for any purpose
large language models (LLMs): artificial intelligence trained on a huge amount of text from books, websites and other sources that can “understand” and generate human language. OpenClaw is the “hands” that execute tasks, and LLMs, like OpenAI’s ChatGPT or China’s DeepSeek, are the “brains”.
auditing: to conduct an official financial inspection of a company or its accounts – to ensure developers and firms are complying with rules
national computing hub: China’s national computing network, which has the world’s second-largest computing capacity behind the United States. This centralised network of massive data centres is the physical backbone of the country’s AI strategy.
Sample answers
News
-
B
-
One way a user in the article uses it is to have their lobster scan Facebook for events and create ticket-buying guides. The AI agent can autonomously manage files, emails and calendars. It performs real-world tasks for users, much like a supercharged digital assistant. (accept all reasonable answers)
-
I would be interested in trying OpenClaw, but I would also be very cautious about using it right now. According to News, OpenClaw is a free, open-source AI agent that connects with large language models to perform real-world tasks like sending messages, managing emails and even booking tickets. It appears to be a convenient tool, as it could save me a lot of time and help organise my life more efficiently, much like how Jay Kwong’s “Puppycat” assists her with daily activities. However, I am concerned about the potential security and privacy risks. Authorities in Mainland China and Hong Kong have warned against using OpenClaw, as it could allow unauthorised users to access private data and control devices. Since it requires users to grant access to sensitive accounts like email and online banking, it could leak personal information and lead to hacking attacks. In conclusion, while OpenClaw seems innovative and useful – combining the “hands” of automation with the “brains” of large language models – I would probably wait until stronger security protections are guaranteed before using it. (accept all reasonable answers)
Infographic
-
After receiving a request from the user, OpenClaw sends a message to the right agent to perform the task, which here is an order on an e-commerce app. It uses information from a large language model to understand which e-commerce app is needed. It adds 12 eggs to the cart, and then a user can check out either manually or grant OpenClaw access to their bank account to make the payment.
-
A user might choose to check out manually to limit the agent’s access to payment applications, to avoid unwanted payments or data breaches.
Issue
-
According to Liu Liehong, security management for AI agents should include giving the tools the smallest level of access needed to do their jobs. It would also require continuous official financial inspections of a company or its accounts to ensure that developers and firms comply with the rules. He noted it would require coordinated action from AI providers, end users and regulators.
-
Liu Liehong, head of China’s National Data Administration, warned that the rapid spread of AI has brought several security and compliance challenges. He identified issues such as copyright disputes over training data and AI-generated content, as well as cybersecurity threats like data poisoning, where attackers tamper with data to change how AI models behave. He also mentioned that AI tools such as OpenClaw, which can control apps and perform real-world tasks, must be managed carefully to prevent unauthorised access and misuse of personal information. These issues reflect global concerns about AI. For example, AI systems often collect large amounts of sensitive data, which can be vulnerable to leaks or hacking. There is also the risk of biased or inaccurate outputs, which could harm users or spread misinformation. As AI becomes more capable of acting independently, the potential for loss of control and ethical misuse grows, making security a top priority. (accept all similar answers)
-
Liu’s “data property rights framework” focuses on defining clear ownership, rights and responsibilities for data use and circulation. It aims to build a system that governs how data is shared and protected, ensuring that all stakeholders, AI providers, users and regulators play a role in maintaining security. In this approach, the burden of safety is shared among multiple parties through oversight, legal structures, and continuous auditing. Han Xinyi’s security-by-design philosophy, on the other hand, places the responsibility primarily on AI developers. He argues that safety and ethical measures should be built directly into AI systems from the start, rather than added after risks appear. This means developers must design AI with robust safeguards against misuse, bias and cyber threats from the ground up. In summary, Liu’s framework addresses data governance and shared accountability, while Han’s philosophy emphasises technical integrity and developer responsibility. Both aim for safer AI, but they differ in where the safety burden lies. Liu’s is distributed, while Han’s is developer-centred.
Illustration
-
The child represents someone who uses the AI agent OpenClaw, depicted in the illustration as the lobster. Hugging means the user supports the tool and regards it affectionately as their personal assistant.
-
The shadow represents the concern of data privacy. OpenClaw requires users to give it broad access permissions to perform real-world tasks, including control over messaging apps, like WhatsApp and Telegram, email accounts, calendars and even online banking. Because OpenClaw is an AI agent, not just a chatbot, it can act on behalf of users to send messages, manage files, and access sensitive information. This creates serious risks if the system is compromised. Authorities in Mainland China and Hong Kong have warned that it could allow unauthorised outsiders to access private data or even take control of users’ devices. By giving OpenClaw permission to handle personal tasks, users may unintentionally expose confidential information, such as passwords, financial details, or personal conversations, to hackers, malicious code or misuse by third parties.
