China Investigation Targets DeepSeek and Moonshot Over Alleged Claude Data Routing

China's Cyberspace Administration Launches Probe Into AI Companies

China's internet regulator has opened an investigation into DeepSeek and Moonshot AI following Anthropic allegations that both companies routed sensitive user data to Claude models without authorization1

1

2

2

. The Cyberspace Administration of China summoned representatives from seven AI companies named in Anthropic's threat intelligence report, but quickly narrowed its focus to DeepSeek and Moonshot based on specific examples detailed in the document1

1

. Regulators have questioned staff from both companies, though no penalties have been decided as the China investigation continues1

1

.

Anthropic Allegations Trigger Regulatory Scrutiny

The investigation stems from a 154-page report published by Anthropic on September 10, which accused seven Chinese labs of feeding user conversations into Claude through model distillation practices2

2

4

4

. Anthropic counted more than 23 million exchanges from Moonshot between May and July, and nearly 300,000 requests in a single ten-day window through 5,380 fraudulent accounts1

1

3

3

. DeepSeek allegedly routed more than 12.1 million exchanges to Claude during 14 days in July3

3

. The data routing involved requests containing names, email addresses, company data, and other sensitive information from hundreds of end users across at least a dozen languages2

2

.

Sensitive User Data at Center of Investigation

What distinguishes this probe from typical AI intellectual property disputes is the nature of the sensitive user data allegedly transmitted. According to Anthropic's report, one example involved a user likely affiliated with the People's Liberation Army who asked Moonshot's Kimi model to analyse surveillance footage tracking a person across hundreds of police cameras in Chengdu, including cameras outside PLA facilities and defense-related institutes1

1

. Moonshot allegedly forwarded this request and the surveillance footage analysis to Claude without informing the user1

1

. DeepSeek reportedly relayed requests involving Chinese police systems and national identification numbers3

3

. These categories fall under China's strictest cross-border data transfer regulations, making the alleged data routing a serious domestic regulatory matter1

1

.

How Model Distillation Works in AI Data Governance

The practice at the heart of these Anthropic allegations is known as model distillation, where smaller AI models use responses from larger, more sophisticated models as training data to improve their own outputs2

2

. Moonshot reportedly showed Claude responses to customers while presenting them as outputs from its own Kimi model1

1

3

3

. This allows companies to bypass the need for massive training data collections by leveraging another model's capabilities. Anthropic described the activity as illicit distillation and argued it was likely inconsistent with data privacy laws and the companies' own terms of service2

2

3

3

.

Timing Complicates Matters for Both AI Companies

The regulatory scrutiny arrives at an awkward moment for both firms. DeepSeek is scheduled to brief the UN Security Council this week on AI risks to international security, positioning itself as a responsible voice on AI governance while facing a domestic investigation into its own data handling practices1

1

3

3

. Moonshot faces different complications as it works towards a Hong Kong listing, where an open regulatory investigation becomes material information that must be disclosed in prospectus documents1

1

. Neither DeepSeek and Moonshot have issued public statements addressing the investigation1

1

3

3

.

U.S.-China Tensions Over AI Intellectual Property Escalate

The investigation adds another dimension to growing U.S.-China tensions surrounding artificial intelligence development and data governance. Just one day before Anthropic published its report, US intelligence agencies including the NSA, FBI and CISA issued an advisory warning about Chinese AI distillation efforts1

1

. China's foreign ministry initially dismissed these accusations as unfounded on September 101

1

. However, twelve days later the Cyberspace Administration had summoned every company named in Anthropic's separate report1

1

. Treasury Secretary Scott Bessent previously indicated that sanctions could follow if evidence verified Chinese model distillation practices3

3

. The timing of China's investigation may serve as an olive branch ahead of meetings between Chinese President Xi Jinping and the Trump administration, where AI regulation is expected to feature prominently in discussions2

2

.

Other Named Companies Face Separate Challenges

While the probe focuses on DeepSeek and Moonshot, other companies identified in Anthropic's report face their own difficulties. The Cyberspace Administration summoned representatives from all seven companies mentioned, including Alibaba, Zhipu, SenseTime, MiniMax and Xiaomi4

4

. Zhipu, which operates as Z.ai, dealt with an unrelated data privacy incident last week when developers discovered its ZCode tool uploading encrypted snapshots of local code repositories without consent1

1

. The company apologized, open-sourced the tool, and commissioned two external assessments1

1

. While these incidents are separate, they raise similar questions about transparency and control when Chinese AI products move user data to unexpected destinations1

1

.

What Industry Watchers Should Monitor

The investigation remains unresolved as Chinese authorities continue examining potential violations3

3

. Any confirmed breaches could reshape privacy controls, model access rules, and safeguards around sensitive prompts across China's AI sector. The outcome may establish precedents for how regulators handle cross-border AI data flows and whether distillation practices violate existing frameworks. For global AI companies, the case highlights the complexity of operating across jurisdictions with different priorities—where Anthropic saw AI intellectual property concerns, Beijing identified potential violations of data sovereignty and user privacy protections. The investigation tests whether international consensus on AI data governance can emerge amid deepening technological competition between major powers.