China Probes DeepSeek, Moonshot Over Claude Data Routing logo

China Probes DeepSeek, Moonshot Over Claude Data Routing

Chinese regulators opened a probe into DeepSeek and Moonshot AI after Anthropic accused seven Chinese companies of secretly routing customer requests through Claude to harvest its outputs, sending Alibaba, Zhipu and MiniMax shares lower.

By the Numbers

7
Companies named by Anthropic
23M+
Moonshot exchanges, May-Jul
12.1M+
DeepSeek exchanges, 14 days
-12.4%
Zhipu stock move
-4.4%
Alibaba stock move
TC
By the Markets Desk
Edited by Trace Cohen · Early-stage VC & angel · Founder, New York Venture Partners
3 min read
ShareXLinkedInEmail

THE RUNDOWN

1

Anthropic's own 154-page threat report, published September 10, named seven China-based companies -- Alibaba, Moonshot AI, DeepSeek, Zhipu, MiniMax, Xiaomi and SenseTime -- alleging they routed customer requests through Claude to harvest its outputs, a practice Anthropic calls illicit distillation.

2

The scale alleged is large: Moonshot sent more than 23 million exchanges through Claude between May and July, and DeepSeek generated more than 12.1 million exchanges in a single 14-day window in July, according to Anthropic's report.

3

China's Cyberspace Administration has sent officials to interview employees at DeepSeek and Moonshot directly, a concrete regulatory step beyond a company statement, and the probe knocked Zhipu down 12.4%, MiniMax down 4% and Alibaba down 4.4% on the news.

4

The dispute complicates the US policy debate over restricting frontier model access to Chinese developers -- Anthropic's own report is now functioning as the evidentiary basis for a separate Chinese government investigation into two of its own accused customers.

TC

The VC Read · Trace's Take

Trace Cohen

The number I'd focus on isn't the stock move, it's the methodology gap -- Anthropic's report is the sole evidentiary source cited by regulators so far, and no independent party has yet verified the 23-million-exchange figure against Moonshot's own logs. If you're diligencing any company that depends on API access to frontier models, the real risk this surfaces is contractual: watch whether Anthropic starts adding harder technical enforcement against session-signature replay, since that's the specific exploit this report describes and it likely isn't unique to the two named companies.

Analysis

Shares of several Chinese AI companies fell on September 23 after Bloomberg and other outlets reported that Chinese regulators have opened a probe into DeepSeek and Moonshot AI over allegations that both companies secretly routed customer data through Anthropic's Claude models, according to Bloomberg and The Information.

What Anthropic Actually Alleged

Anthropic published a 154-page threat intelligence report on September 10 accusing seven China-based AI companies -- Alibaba, Moonshot AI, DeepSeek, Zhipu, MiniMax, Xiaomi and SenseTime -- of routing customer requests through Claude to harvest its outputs for training their own models, a practice the report calls illicit distillation. The scale Anthropic describes is large and specific: Moonshot allegedly sent more than 23 million exchanges through Claude between May and July, and relayed almost 300,000 customer requests through 5,380 fraudulent accounts in a single 10-day window. DeepSeek's operation, per the report, generated more than 12.1 million exchanges in just a 14-day stretch in July. Anthropic says the intercepted prompts contained names, email addresses, company data and other sensitive information belonging to hundreds of end users across at least a dozen languages, including requests it says came from a Chinese police surveillance system project.

DeepSeek's operation, per the report, generated more than 12.1 million exchanges in just a 14-day stretch in July.

Beijing's Response

China's Cyberspace Administration has sent officials to interview employees directly at both DeepSeek and Moonshot, a concrete investigative step rather than a routine statement, according to South China Morning Post. The inquiry is reportedly focused on whether the companies committed security or data-handling violations under Chinese law, separate from -- but triggered by -- Anthropic's own allegations of intellectual-property theft against them.

The Market Reaction

Zhipu (also known as Z.ai) fell 12.4%, MiniMax dropped 4%, and Alibaba -- which Pulse has covered extensively on its own AI buildout -- declined 4.4% on the news, with Xiaomi and Tencent also trading lower amid broader concern that Beijing could tighten scrutiny across the domestic AI sector. That is a meaningful one-day move for companies at this scale, and it reflects investor uncertainty about how far the regulatory response could extend beyond the two named companies -- five other firms remain named in Anthropic's report but not yet confirmed as subjects of the CAC probe.

What This Reveals, And What It Doesn't

The episode is genuinely unusual: a US AI lab's own competitive-intelligence report is now functioning as the evidentiary trigger for a foreign government's regulatory investigation into companies that lab considers rivals. That dynamic complicates the broader US policy argument that restricting frontier-model access to Chinese developers is necessary because those developers can't be trusted with it -- Beijing's response here suggests Chinese regulators also don't want their own companies improperly extracting value from a foreign competitor's model, which is a different concern than the containment argument Washington has made.

What's not yet established is whether Anthropic's technical claims hold up to independent scrutiny, or whether DeepSeek and Moonshot will contest the specific exchange counts and methodology behind them. Neither company has issued a detailed public rebuttal as of this writing, and a Cyberspace Administration probe determining a security violation is a different bar than confirming Anthropic's distillation allegations are accurate in full.

The number that will matter more than today's stock moves is whether the CAC's findings, once public, corroborate Anthropic's specific numbers -- that would turn a single lab's competitive report into an independently verified account of how the illicit-distillation problem actually works at scale.

ShareXLinkedInEmail

Key Sources

2 sources

Reported by Bloomberg · Analysis by Value Add Pulse.

← Back to Pulse

THE WIRE in your inbox— Tech, startup & VC news with Trace's take. Free, no spam.