China’s Top Spy Chief Warns A.I. Is a Threat to Party Rule

The New York Times | Lily Kuo

On September 14, 2026, Chen Yixin, minister of the Chinese Ministry of State Security, stated that artificial intelligence represents a direct threat to the rule of the Chinese Communist Party and urged an increase in party control as well as more stringent regulatory supervision. In an article published in China Cyberspace, Chen pointed out the dangers posed by sophisticated foreign models such as Claude Mythos and GPT-5.5-Cyber, giving as examples the use of deepfakes for propaganda, cyberattacks on critical infrastructure, and hostile intelligence activities.

Foreign intelligence agencies are actively using algorithm-based tools to attack domestic networks, and there are still serious breaches of information security. It has recently come to light that the Chinese company Moonshot AI directed queries involving Chinese military video surveillance data to the American Claude servers. Moreover, the use of American military AI during the conflict between the U.S. and Israel over Iran shows just how automated sensing and algorithmic targeting have fundamentally changed modern warfare on the battlefield. In order to meet these developing threats, Beijing is placing a higher priority on data sovereignty and state security than on international agreements concerning AI safety ahead of the September 24 bilateral summit in Washington.

Comment

The official classification by the Ministry of State Security of foreign frontier language models as cyber vectors shows a shift in strategy towards counter-algorithmic network defence. Since it regards systems like GPT-5.5-Cyber as hazards to the infrastructure, Beijing considers the use of AI to generate payloads to be an immediate threat to the system as a whole. This was proven by the case in which automated tools for exploiting vulnerabilities targeted the architecture of the WeChat application in order to bypass its authentication controls.

In a mechanical way, the use of automated vulnerability detection cuts down the time needed to carry out zero-day attacks on state firewalls. Since query transfers go around domestic security measures—as shown by Moonshot AI directing surveillance telemetry to Claude servers—unencrypted intelligence data becomes open to remote interception. As a result, the Ministry of State Security's responsibility for cyber defence is now pushing for air-gapped API architectures in order to isolate China's critical information networks from external foundation models.

Strategic Question for Discussion

When domestic companies such as Moonshot AI are routing their operational telemetry through foreign platforms like Claude, which of the two presents a greater systemic challenge to a state's cyber defence—uncontrolled data exfiltration or automated payload injection?

The evidence currently available suggests that data exfiltration is the more urgent structural threat since telemetry data that is not encrypted can reveal operational intelligence without any need for gaining access to the network. Although there are serious tactical risks involved in carrying out automated payload injections on platforms such as WeChat, such actions still require existing exploitation pathways. As a result, the defenses at the perimeter are likely to focus on closely examining API traffic rather than using wide-scale algorithmic filtering.

Share your assessment in the comments below.

To read in details click here