Yahoo Finance, carrying a Stocktwits report citing The Information, says China’s Cyberspace Administration has reportedly examined DeepSeek and Moonshot AI after Anthropic alleged the companies routed exchanges through its Claude models. The report is significant, but the source material reviewed for this article does not include a public CAC statement confirming its central regulatory claim.
The Yahoo Finance report said, citing The Information, that CAC was reportedly examining whether potentially sensitive data tied to Chinese police, military, or state-linked companies had reached U.S. AI models through DeepSeek’s and Moonshot’s use of Claude. Anthropic’s September threat-intelligence report supplies the company’s own account of alleged routing and distillation activity.
The proper wording is therefore narrow. Yahoo Finance, citing The Information, reported CAC scrutiny. Anthropic alleged that it identified particular activity. Neither source independently establishes a public CAC finding, a legal violation, a penalty, or an outcome.
Reported CAC Review and Its Public Evidence
According to Yahoo Finance’s account of The Information, CAC first summoned representatives of the seven firms included in Anthropic’s account, before the reported scrutiny narrowed more specifically to DeepSeek and Moonshot. Yahoo listed the companies as DeepSeek, Moonshot AI, Alibaba, Zhipu, SenseTime, MiniMax, and Xiaomi.
The reporting does not establish that every company was reviewed in the same way. It does not describe a publicly stated legal basis, subpoena, penalty, enforcement order, or conclusion. It also does not show that DeepSeek or Moonshot admitted wrongdoing.
The absence of an official CAC statement is not evidence that the reported review did not occur. It simply defines the level of certainty available to readers. Reported CAC review and reported scrutiny are accurate. CAC proved or CAC found would go beyond the sources.
The source chain also matters. The article on Yahoo Finance is a Stocktwits report by Prabhjote Gill. It attributes the CAC account to The Information. The Information’s original wording was not separately available in the materials reviewed here, so the correct attribution is not merely The Information reported, but Yahoo Finance, citing The Information, reported.
That clarity keeps a potentially consequential story from becoming more definite than its evidence permits. It also separates the alleged regulatory action from Anthropic’s own claims about model use.
Anthropic’s Distillation Allegations and Exchange Counts
Anthropic alleges that DeepSeek, Xiaomi, and Moonshot sent exchanges involving their own models and users to Claude, then used Claude responses as training material for distillation. That is Anthropic’s allegation and account of activity detected in its systems. It is not an independent adjudication of the conduct.
Anthropic also says some exchanges included sensitive information from individual users, multinational companies, and state-affiliated actors. Yahoo Finance’s account puts Moonshot’s alleged May-to-July total above 23 million exchanges sent to Claude, and says DeepSeek’s alleged 14-day July total exceeded 12.1 million.
Those figures should not be converted into a number of people, unique prompts, complete conversations, or confirmed sensitive records. They are exchange totals reported in connection with Anthropic’s account. The presence of a large number does not establish that every exchange had the same content or sensitivity.
One stated example concerns engineers working on a municipal Chinese Public Security Bureau’s case-management system. Anthropic alleged that DeepSeek relayed requests containing data used to compare a person’s movements with police records and national identification numbers. The example is source-specific and remains an Anthropic allegation, not an independently verified incident description.
The available reporting does not answer every technical question about the exchanges. It does not establish that a single pathway was used in every case. It also does not establish that every DeepSeek or Moonshot customer request was routed to Claude. The source-bounded point is limited to what Anthropic said it identified and what Yahoo Finance reported.
A Separate U.S. Government Advisory Provides Broader Context
A September 8 CISA, NSA, and FBI advisory separately alleged that China-based AI companies, including DeepSeek and Moonshot, conducted industrial-scale distillation campaigns against U.S. frontier models. The advisory said the activity used multiple pathways, including APIs, remote cloud providers, third-party aggregators, and gray-market transfer stations.
That advisory is important context, but it cannot independently verify the reported CAC action. It also does not establish which route was used in every DeepSeek or Moonshot exchange referenced by Anthropic. The government advisory and the Yahoo Finance report should remain separate pieces of evidence with different claims and different sources.
Anthropic’s report says the activity it covered was disrupted between December 2025 and August 2026 and involved seven harm areas. That time period describes Anthropic’s broader research scope, not a date for any CAC action. The Yahoo Finance report is dated September 22, while Anthropic’s report is dated September 10 and the U.S. advisory is dated September 8.
EastFrontier’s archive includes earlier reporting on China’s Claude Code security alert and Moonshot-related distillation allegations. Those prior articles do not confirm the current reported CAC scrutiny.
The conclusion should stay with the evidence. Yahoo Finance, citing The Information, reports that CAC is scrutinizing DeepSeek and Moonshot. Anthropic alleges routing and distillation activity. A U.S. advisory alleges broader industrial-scale distillation by China-based companies. What remains absent from the reviewed material is a public CAC statement, a company response, or a formal finding. Until those appear, each claim should carry the attribution that belongs to it.