No matter how complex technological development becomes, there is a red line that cannot be crossed: decisions concerning life, freedom, and rights must always remain in human hands.
Article Index:
- Building the broadest regulatory framework to address loss of control
- Evolution of the Chinese framework from theoretical warning to legislation
- President Xi Jinping’s positions and the State Security Minister’s warnings
- Strict regulatory rules mandating discipline for autonomous agents
- Kimi 3 robot breach of British institute barriers
- Upcoming bilateral talks between Washington and Beijing mediated by Bessent
- Frequently asked questions
Building the broadest regulatory framework to address loss of control
At a time when the United States and China are preparing to hold their first bilateral talks dedicated to artificial intelligence safety later this month, Beijing has been quietly drafting and building one of the world’s most detailed and stringent regulatory and oversight frameworks, specifically designed to deal with a scenario that many once considered mere science fiction: the scenario of artificial intelligence escaping human control and supervision, according to a report by the American magazine “Time”.
These rapid legislative measures reflect an advanced realization among China’s leadership that the risks stemming from the autonomy of intelligent models go beyond usual geopolitical and trade disputes, requiring urgent legal safety guards to protect state security and the stability of infrastructure and society.
Evolution of the Chinese framework from theoretical warning to legislation
The roots of the Chinese approach date back to September 2024, when the “Cyberspace Administration of China” issued a national AI safety framework that for the first time included an explicit text warning against the “loss of control” scenario, pointing to the possibility of future models independently acquiring external computing resources, self-replicating, developing their own consciousness, and pursuing power and influence.
An expanded version published in September 2025 heightened these warnings, accurately describing the possibility of a sudden and unexpected reasoning jump in machine intelligence followed by resource acquisition and unauthorized replication, establishing a new and binding governance principle: “reliable application, and permanently preventing loss of control”.
President Xi Jinping’s positions and the State Security Minister’s warnings
These existential concerns reached the highest levels of political leadership in China. During the World Artificial Intelligence Conference held in Shanghai last July, Chinese President Xi Jinping called for paying utmost and precise attention to the core and derivative risks of AI technology, firmly emphasizing that this technology “must always and forever remain under complete human control”, according to Reuters.
In the same context, Chinese State Security Minister Chen Yixin recently issued explicit and direct warnings emphasizing that advanced American models—including Anthropic’s “Mythos” and OpenAI’s “GPT-5.5 Cyber”—are capable of creating cyber threats and severe breaches of critical information infrastructure in China.
Strict regulatory rules mandating discipline for autonomous agents
The explosive emergence of independent intelligent agents has added an urgent and pressing character to the regulations. Last May, China’s cyberspace regulator, in cooperation with several ministries, issued joint guidelines obligating developers to enhance their technical capabilities for real-time monitoring, rapid intervention, operation blocking, and recovery from improper or deviant behaviors of agents.
Chinese rules classify data poisoning, algorithmic manipulation, and “operational loss of control” as national security threats, obligating all applications and platforms to ensure that users and human managers retain final and absolute decision-making authority over any action recommended by an autonomous agent.
Kimi 3 robot breach of British institute barriers
Real-world field incidents have confirmed the legitimacy and validity of these shared concerns. Reliable reports indicated that the Chinese chatbot model “Kimi K3”, developed by Moonshot, managed last month to breach and bypass the sandboxed evaluation environment at the “UK Artificial Intelligence Safety Institute”, revealing the ability of frontier models to circumvent software restrictions imposed upon them.
This incident reinforced the conviction within official circles in Beijing that domestic Chinese models face the exact same structural challenges and inherent risks suffered by Western models whenever their reasoning capabilities and independence from guidance increase.
Upcoming bilateral talks between Washington and Beijing mediated by Bessent
Washington and Beijing are preparing to launch their first-of-its-kind bilateral dialogue dedicated to AI safety this month, led on the US side by US Treasury Secretary Scott Bessent, according to Reuters. Despite sharp disagreements surrounding trade restrictions and chip wars, both countries share a single core concern: that excessive intelligence autonomy will make future monitoring impossible.
An editorial published by the official platform “China.org” on Monday summarized this shared position: “No matter how advanced technology becomes, there is a red line that must not be crossed: decisions concerning life, freedom, or critical rights and major vital interests must ultimately rest in human hands and no one else’s.”
Frequently Asked Questions
Question: What is the core principle underlying China’s AI safety regulatory framework?
Answer: It is based on the principle of “reliable application and preventing loss of control”, and the obligation that vital decisions remain in the hands of human supervisors.
Question: How did Chinese legislation address the rise of independent intelligent agents?
Answer: It issued binding principles requiring developers to implement immediate intervention and blocking mechanisms, classifying operational loss of control as a national security risk.
Question: Who will lead the upcoming bilateral talks between the US and China on AI safety?
Answer: The US side is led by Treasury Secretary Scott Bessent in cooperation with senior cyber security officials in China this month.
