China's Cyberspace Administration Investigates DeepSeek and Moonshot for Alleged Claude Data Leaks

China’s Cyberspace Administration has formally opened an investigation into two of the country’s most prominent AI labs—DeepSeek and Moonshot—after Anthropic, the U.S. creator of the Claude model, accused them of secretly routing millions of user queries to its servers for illicit model distillation. The regulator summoned representatives from all seven firms named in Anthropic’s September 10 threat-intelligence report, but the focus quickly narrowed to DeepSeek and Moonshot, the two entities with the largest alleged exchange volumes.

1. The core allegation and its technical nuance

Anthropic’s report, filed on September 10, details a practice known as “distillation,” where a smaller model is trained on the outputs of a larger, often more capable, model. While distillation itself is a legitimate technique, Anthropic alleges the Chinese labs used fraudulent accounts to siphon user queries to Claude, effectively outsourcing inference without paying for the service. The numbers are stark: Moonshot allegedly routed 23 million exchanges through 5,380 fake accounts, while DeepSeek generated 12.1 million exchanges within a 14-day window in July. These figures suggest systematic, large-scale data extraction rather than isolated incidents.

2. Timing and strategic exposure

The investigation arrives at a delicate moment for both firms. DeepSeek is slated to brief the United Nations Security Council on AI risks this week, positioning itself as a responsible stakeholder in global AI governance. Simultaneously, Moonshot has confidentially filed for a $3 billion IPO in Hong Kong, a move that would make it one of the largest Chinese AI listings to date. The regulatory spotlight could delay or even derail the IPO, as Chinese securities regulators typically require clear compliance with data-security statutes before approving such offerings.

3. Regulatory backdrop and enforcement posture

China’s Cyberspace Administration (CAC) has intensified its oversight of AI after the State Council’s “Regulation on the Administration of Generative AI Services” took effect earlier this year. The CAC’s mandate now includes safeguarding “national security data,” which explicitly covers police, military and state-linked corporate information. By targeting DeepSeek and Moonshot, the CAC is effectively testing the enforcement of these new rules against the most technically sophisticated domestic AI players.

4. Potential market ramifications for digital assets

The investigation could reverberate across crypto markets in several ways:

• Capital flight risk – Investors may pull funds from AI-related token projects that rely on Chinese cloud infrastructure, fearing sudden service interruptions.
• Liquidity squeeze – If the CAC mandates data-localisation or bans cross-border model training, platforms that tokenise AI outputs could see reduced on-chain activity, impacting volume metrics on DEXs.
• Compliance cost surge – AI startups may need to implement robust data-audit trails, driving up operational expenses and potentially reducing the attractiveness of token-based fundraising.

These dynamics echo the broader trend where regulatory actions in one jurisdiction cascade into crypto-related market adjustments, as seen after the U.S. SEC’s crackdown on unregistered securities offerings.

5. Institutional response and risk mitigation

Institutional investors with exposure to Chinese AI ventures should reassess their risk matrices. Key steps include:

1. Data-flow audit – Verify that any AI model training pipelines do not inadvertently transmit user data to foreign servers.
2. Legal review – Ensure compliance with the CAC’s data-security provisions, especially regarding “sensitive information” definitions.
3. Diversification – Consider reallocating capital toward AI projects operating within jurisdictions with clearer regulatory frameworks, such as the EU’s AI Act-aligned ecosystems.

6. What remains unconfirmed

Anthropic’s report is the primary source of the allegations; the CAC has not publicly disclosed any technical findings or evidence of actual data leakage. Moreover, the investigation’s scope—whether it will expand to the other five labs named (Alibaba, Zhipu, MiniMax, SenseTime, Xiaomi)—remains unclear. Until the CAC releases a formal statement, the claims should be treated as allegations rather than proven misconduct.

7. Corroborating coverage and broader context

The story aligns with coverage from Crypto News, which highlighted Anthropic’s broader concerns about illicit model distillation by Chinese firms. While that article focuses on price-target predictions, it underscores the industry-wide anxiety over data-privacy violations and the potential for regulatory backlash.

8. Looking ahead – watch points for operators

• Regulatory filings – Monitor any CAC releases or court filings related to the probe; they will clarify enforcement thresholds.
• IPO timeline – Track Moonshot’s Hong Kong listing schedule; a postponement would signal heightened compliance demands.
• UN briefing outcomes – DeepSeek’s UN Security Council presentation may reveal how the firm frames its data-governance practices on the world stage.
• Cross-border AI collaborations – Expect tighter contracts for any partnership that involves routing queries to foreign models, especially those hosted in the U.S.

By staying attuned to these signals, institutional participants can better navigate the evolving regulatory landscape that sits at the intersection of AI, data privacy, and digital-asset markets.

---

This analysis is based on reporting from Decrypt and corroborated by cryptonews.com. All figures are drawn from the original Anthropic threat-intelligence report and CAC statements as cited.

Related coverage

• LINK Price Prediction: Anthropic Claude AI Forecasts 300% Surge by 2027
• TRON Inc.’s TRX Funding Shifts After UK Sanctions on HTX: Regulatory and Market Implications for TRON Inc. TRX

Explore more on this topic

• More ai-regulation coverage