China’s Security Chief Warns of Six AI Risks Amid Reported Military Data Probe

China’s state security chief has warned that artificial intelligence poses growing risks to political control, cybersecurity and military operations, as industry sources claim authorities are examining whether sensitive military and police data was exposed through AI platforms.

The concern follows a report from U.S. artificial intelligence company Anthropic showing that Chinese AI providers routed some users’ requests to its Claude model without their knowledge. The material included surveillance information from a user assessed as likely connected to the People’s Liberation Army (PLA), as well as data involving a Chinese police system.

Chinese authorities have not publicly confirmed that an investigation is underway.

Chen Yixin outlines six AI risks

On Sept. 13, China Cyberspace magazine published a signed article by Chen Yixin, head of the Chinese Communist Party’s Ministry of State Security.

Chen described AI as a “main battlefield of global technological competition” and divided the risks facing China into six categories: threats to political security, increasingly sophisticated cyberattacks, large-scale theft and disclosure of sensitive information, technological monopolies, disruption of social governance and changes to modern warfare.

For the best of our weekly content！

You are now signed up for our newsletter

Check your email to complete sign up

The article said China had more than 500 million AI-product users as of June 2026, with daily token usage exceeding 140 trillion. Chen accused foreign adversaries and other actors of using deepfakes, AI-generated content and automated influence networks to threaten China’s “political security, institutional security and ideological security.”

Chen also warned that Chinese users handling sensitive material with overseas AI products could cause “large-scale leakage” of domestic data. He called for stronger monitoring, new national and regional data pools, and accelerated development of Chinese computing chips, software frameworks and AI applications.

The X post below highlighted Chen’s data-leak warning following the article’s publication:

Chen’s proposed response placed the CCP at the center of AI governance and called for key technologies to remain under domestic control. On military applications, he argued that increasingly precise AI-assisted sensing, judgment and targeting would help determine which side gains the battlefield advantage.

Anthropic says Chinese firms routed user data to Claude

The warning followed Anthropic’s Sept. 10 threat-intelligence report, which accused seven China-based AI companies—Alibaba, Moonshot AI, DeepSeek, Zhipu, Xiaomi, SenseTime and MiniMax—of improperly accessing Claude outputs to improve their own models.

Anthropic said Moonshot, developer of the Kimi chatbot, “silently forwarded customer requests to Claude” and then displayed Claude’s responses as if they had been generated by Kimi.

During one 10-day period, Moonshot allegedly relayed nearly 300,000 customer requests through 5,380 fraudulent accounts, most of which appeared to be based in Singapore and Japan. Anthropic attributed more than 23 million exchanges to Moonshot between May and July.

Some requests contained sensitive information. In one case, a user Anthropic assessed as “likely affiliated with the PLA” uploaded surveillance data from hundreds of cameras in Chengdu, including cameras near PLA facilities, institutes connected to China Electronics Technology Group Corp. and a major state-owned enterprise.

The user reportedly believed the information was being processed by Kimi and did not know it had been forwarded to Anthropic.

Anthropic also said an engineer at a major Chinese state-owned enterprise submitted internal code and active corporate credentials through Kimi. Separately, DeepSeek allegedly forwarded requests involving a case-management system being developed for a municipal Public Security Bureau. The system compared individuals’ movements with police records using national identification numbers.

DeepSeek routed more than 12.1 million exchanges to Claude during a 14-day period in July, according to the report.

Anthropic’s findings indicate that at least some sensitive information reached a U.S. AI system through Chinese platforms rather than through users deliberately accessing Claude.

In response to broader U.S. accusations of illicit model distillation, China’s Commerce Ministry said the allegations “lacked factual basis and legal grounding”. The ministry described distillation as a “neutral technical method in nature” and accused Washington of attempting to suppress Chinese competition.

Sources report internal investigation

A Chinese AI engineer identified only as Mr. Chu said authorities were investigating whether military personnel had improperly used overseas AI products, according to Chinese-language reporting by The Epoch Times republished by Apollo News.

Chu said his company tightened its internal controls after Anthropic released its findings and that the incident was being treated as a “family scandal.”

“If Anthropic’s disclosures are true, this clearly involves unauthorized operations by personnel within the military,” he said.

A second source, identified by the pseudonym Hu Chenfeng and described as an insider in China’s internet-management system, said authorities were attempting to determine how much sensitive information had entered the systems of U.S. AI companies.

Hu claimed personnel associated with the Chinese military, police and design institutes had used overseas AI tools for more than two years. Investigators now needed to establish “which organizations were involved” and “whether the data is still being retained,” he said.

The sources’ claims have not been independently verified. As of publication, China’s Cyberspace Administration, Ministry of Industry and Information Technology, Ministry of Public Security and Ministry of National Defense had not publicly announced an investigation.