China's CAC Investigates DeepSeek and Moonshot Over Alleged Routing of User Data to Anthropic's Claude
Summary
- • China's CAC investigates DeepSeek and Moonshot AI over alleged user data leaks to Anthropic via Claude
- • Anthropic's Sept. 10 threat intelligence report alleged 35M+ Claude exchanges routed by the two firms
- • CAC questioned staff; probe narrows from 7 Chinese labs named in Anthropic's report to two firms
- • DeepSeek routed 12.1M exchanges; Moonshot routed 23M+ including via 5,380 fraudulent accounts
Details
CAC opens formal investigation into DeepSeek and Moonshot AI
China's Cyberspace Administration (CAC) launched an investigation into DeepSeek and Moonshot AI over alleged covert routing of sensitive user data to Anthropic's Claude models. Staff at both firms have been formally questioned.
Triggered by Anthropic's 154-page threat intelligence report (Sept. 10, 2026)
Anthropic released a 154-page threat report on September 10, 2026, alleging multiple Chinese AI labs used fraudulent accounts to route user data—including police, military, and state-linked corporate information—through Claude's API without authorization.
Moonshot routed 23M+ Claude exchanges May–July 2026 via fraudulent accounts
Anthropic alleged Moonshot routed over 23 million exchanges to Claude between May and July 2026, including approximately 300,000 customer requests over a single 10-day period via 5,380 fraudulent Anthropic accounts.
DeepSeek routed 12.1M+ Claude exchanges over 14 days in July 2026
Anthropic alleged DeepSeek routed over 12.1 million exchanges through Claude over a 14-day period in July 2026, making it the second-largest alleged offender after Moonshot.
CAC narrowed focus to two firms from seven labs initially named
The CAC first summoned representatives from all seven Chinese labs named in the Anthropic report (DeepSeek, Moonshot, Alibaba, Zhipu/Z.ai, SenseTime, MiniMax, Xiaomi) before focusing its active probe on DeepSeek and Moonshot.
Potential violations of China's cross-border data transfer rules
Regulators are examining whether the alleged data routing constitutes a violation of China's cross-border data transfer regulations. No formal sanctions have been announced as of September 22, 2026.
Sensitive data routed included police, military, and state-linked corporate records
Anthropic's report alleged the routed user data included sensitive state-linked content—police records, military information, and data from corporations connected to Chinese government entities.
Routing mechanism used fraudulent Anthropic Claude API accounts
The alleged method involved creating thousands of fraudulent Anthropic API accounts to silently forward user queries from Chinese AI platforms to Claude models, bypassing direct commercial API agreements.
Details compiled via Grok live web research
Additional specifics on the Anthropic threat report, exchange volumes, and CAC probe scope sourced via Grok live web research from AI Weekly, The News International, TechTimes, and Stocktwits coverage.
Primary source: The Information (via TechMeme aggregator). Supplemented by GROK RESEARCH DIGEST citing AI Weekly, The News International, TechTimes, and Stocktwits.
What This Means
This story is simultaneously a data security scandal, a geopolitical confrontation, and a domestic regulatory enforcement action in China. Chinese AI companies allegedly used Anthropic's own infrastructure to serve users—possibly including sensitive state-linked clients—while China's regulator now investigates those same companies for the resulting data exposure under its cross-border data rules. The CAC's willingness to act on a U.S. company's allegations against Chinese firms signals how seriously Beijing views the data sovereignty issue. For Anthropic, the threat intelligence report positions it as both a victim of unauthorized API access and a transparency actor in AI security—a dual role that may shape how AI companies approach competitor misuse reporting going forward.
Sentiment
Concerned focus on data sovereignty violations and ongoing regulatory probes
“SCOOP: China’s internet regulator investigates DeepSeek, Moonshot for potential data leaks to Anthropic, which earlier alleged both companies of routing sensitive user data to Claude models.”
“news: Chinese authorities are investigating Anthropic allegations that AI firms like Moonshot and DeepSeek may have facilitated leaks of 'sensitive Chinese military, police and state-owned corporate data to the U.S., violating Chinese data security laws and regulations.'”
“China is now investigating DeepSeek and Moonshot after Anthropic accused them of secretly forwarding some requests to Claude. ... investigation is ongoing, and nothing is proven yet. China’s concern is if sensitive domestic police and military-related data was sent to a US company without the users knowing.”
“Anthropic published a report earlier this month outlining how seven Chinese companies were using Claude illicitly on a widespread scale. Shortly after, the Cyberspace Administration of China summoned each firm’s representatives. The regulator later zeroed in on DeepSeek and Moonshot.”
Split
Broad consensus on the facts and seriousness of data routing (~85/15 agreement on regulatory implications vs. unproven claims).