U.S.-China AI rivalry burns into the data security domain! According to a report by Yahoo Finance, China’s internet regulator has launched an investigation into DeepSeek and Moonshot. The trigger is an accusation by U.S. AI giant Anthropic, alleging that these two Chinese companies, without authorization, secretly routed large numbers of user requests (including sensitive data from the police and the military) to its Claude model for processing. The incident happened on the eve of the meeting between Chuan and Xi, drawing intense market attention. (Background: the most terrifying AI experiment—commanding a robot to stab a baby doll with a knife.) (Additional background: a mysterious buyer snapped up 50 tons of obscure old books at Japanese bookstores; the industry suspects AI is collecting them.) As global competition in artificial intelligence (AI) heats up, the technological standoff between the U.S. and China—along with data security concerns—has again come to the forefront. According to Yahoo Finance’s report on September 22, China’s internet regulator, the Cyberspace Administration of China (CAC), reportedly has launched a deep investigation into the well-known Chinese AI startups DeepSeek and Moonshot. The spark for this controversy traces back to a threat intelligence report issued by U.S. AI giant Anthropic in September. The report alleges that several Chinese AI companies secretly used user requests and sensitive data and forwarded them to their Claude models for processing. Anthropic names seven Chinese firms: millions of requests secretly “resold/forwarded” to Claude According to Anthropic’s report, seven Chinese AI companies are accused of large-scale, unauthorized use of the Claude model, including DeepSeek, Moonshot, Alibaba, Zhipu, SenseTime, MiniMax, and Xiaomi. As exclusively disclosed by The Information, the CAC has already summoned representatives from these seven companies in succession, and the current investigation focus is concentrated on DeepSeek and Moonshot. Anthropic makes stern allegations that these two companies not only used Claude to develop their own models, but also directly “forwarded” users’ conversation requests to Claude for processing. Data shows that between May and July 2026, Moonshot forwarded more than 23 million dialogue exchanges; while DeepSeek, in just 14 days in July, forwarded as many as 12.1 million. This means that many Chinese users believe they are using domestic large models, but behind the scenes it is actually Claude from the United States that is doing the work. Involving highly sensitive military and police data, the investigation flares up right before the Chuan-Xi summit, stirring political nerves What worries China’s regulators most is the risk of state-level data leaks. The CAC is closely examining whether sensitive information from Chinese police, military, and entities affiliated with the state flowed into U.S. AI models during the forwarding process. Anthropic, in its report, provides a specific example: an engineer, while developing a case-management system for a city public security bureau in China, forwarded requests to Claude through DeepSeek—only to find that the requests included highly sensitive information such as comparisons of personal movement trails, police records, and even citizens’ identity card numbers. Notably, the timing of this investigation coincides with the eve of a meeting between U.S. President Donald Trump and China’s President Xi Jinping. Reports say the two sides are discussing the possibility of establishing an “AI incident reporting and notification mechanism.” Against this geopolitical backdrop, the investigation into DeepSeek and Moonshot is not merely a dispute between companies over copyright and usage rules; it also highlights a trust crisis between the U.S. and China in AI development and the tug-of-war over national security safeguards. Related report Trueo withdraws from Base back to Ethereum: Mainnet is the best foundation for prediction markets Why I’m not worried that AI will destroy humanity—and you don’t need to worry either? Tomorrow Trump will personally go to the airport to welcome Xi Jinping: Rare earth supply, AI chip export controls, and Taiwan arms sales competing at the same table Morgan Stanley report: The AI semiconductor market will reach $753 billion by 2030, naming 23 Taiwan stocks that benefit from it "Caught for secretly wiring into Claude? China’s CAC investigates DeepSeek and Moonshot; Anthropic controls sensitive data leaks" This article was first published on Dongqu BlockTempo (Dongqu—most influential blockchain news media).