C
hina is stepping up efforts to address the possibility that increasingly powerful artificial intelligence systems could escape human control, as concerns over the safety of advanced AI grow in both China and the United States.
Warnings from researchers at leading US AI developer Anthropic that advanced models could eventually act beyond human control have drawn attention in Beijing. China and the United States are the two major forces driving the development and adoption of frontier AI, but they have developed different approaches to managing the technology’s risks.
While debate in the United States has increasingly focused on whether advanced AI could pose an existential threat to humanity, Chinese policymakers have generally treated AI as a powerful but manageable technology. Beijing’s approach relies heavily on regulation, technical standards, security assessments and state oversight rather than independent monitors operating inside AI companies.
Experts say Chinese and US researchers broadly recognize many of the same dangers but differ in how those risks are framed and addressed.
China’s approach is also influenced by the structure of its AI industry. Chinese developers have increasingly promoted open-weight models, whose underlying parameters can be downloaded, inspected and modified. Leading US companies such as Anthropic and OpenAI generally keep those specifications private.
The difference has created competing arguments over AI safety. Supporters of open-weight models say greater access allows cybersecurity researchers to inspect systems and adapt them for defensive purposes. Critics warn, however, that models that can be downloaded and modified can also be redistributed or altered with fewer safeguards.
One of Beijing’s main concerns is the emergence of AI agents capable of operating with greater independence than conventional chatbots.
A policy issued in May by China’s cyberspace regulator, economic planner and industry ministry identified “operational loss of control” as a security risk for AI agents. Such systems can plan and execute multiple tasks with limited human intervention.
The policy requires developers to strengthen their ability to detect, intervene in, block and recover from improper agent behavior. It also identifies risks including data poisoning, manipulation of algorithms and vulnerabilities in AI systems.
The rules emphasize that users should be informed about autonomous decisions made by AI agents and should retain final decision-making authority.
China has also begun drafting a mandatory national standard specifically addressing AI-agent safety. According to Brian Tse, founder and CEO of AI safety and governance research group Concordia AI, the proposed standard would be the first of its kind.
Chinese regulators have increasingly warned about the possibility that frontier AI systems could circumvent restrictions placed on them. Wang Lihong, a senior official at China’s cyberspace regulator, said particular vigilance was necessary against models that could escape sandbox environments, bypass safety boundaries or attack external production systems.
China’s concerns extend to foreign AI systems as well. Chinese State Security Minister Chen Yixin recently warned that advanced US models, including Anthropic’s Mythos and OpenAI’s GPT-5.5-Cyber, could pose serious risks to China’s critical information infrastructure.
At the same time, incidents involving Chinese AI models have demonstrated that the risks are not limited to US-developed systems.
Researchers said Moonshot’s Kimi K3 recently bypassed a testing sandbox operated by the UK’s AI Security Institute. The incident highlighted concerns that Chinese models could also evade restrictions designed to limit their access to systems and their ability to take actions.
Recommended
China’s concern over AI escaping human control is not new. Beijing explicitly incorporated such a scenario into an AI safety framework released in September 2024 under the guidance of the Cyberspace Administration of China.
The framework warned that future AI systems could potentially obtain external resources autonomously, replicate themselves, develop self-awareness and seek external power. Such developments, it said, could create a risk that AI systems would compete with humans for control.
An expanded framework issued in September 2025 sharpened the warning. It described the possibility that AI could experience a sudden and unexpectedly large increase in intelligence before acquiring resources, replicating itself and seeking power.
The updated framework also introduced the governance principle of “trusted application, preventing loss of control.” An expert interpretation published on the cyberspace regulator’s website said the principle was designed to address risks to human survival and development, including a possible scenario in which AI could effectively “break loose.”
Despite these warnings, China has not adopted the same approach as some Western AI-safety advocates who have called for companies to slow or pause development of the most advanced models until stronger safeguards are established.
Instead, Beijing has pushed aggressively for AI adoption across industries, viewing the technology as an important engine for economic growth. Since early this year, China has promoted the integration of AI into sectors across the economy.
However, the government has demonstrated that it can delay deployment when officials believe regulation has not caught up with technological development. In 2023, Chinese companies delayed the release of some AI chatbots while regulators finalized rules governing generative AI services. Major products were released after the regulations came into force in August.
The contrasting approaches in China and the United States are likely to become increasingly important as the two countries continue competing over advanced AI technology while also discussing AI safety.
For Beijing, the central challenge is to expand the use of AI while maintaining sufficient oversight to prevent systems from behaving in unpredictable or uncontrollable ways. As AI models become more capable and autonomous, China’s regulatory framework is likely to face growing pressure to keep pace with the technology it is designed to govern.
(Source: Reuters)
Stay informed by subscribing to
our weekly newsletter.
