DeepSeek and Moonshot Accused of Secretly Routing Over 300,000 Users to Claude
Anthropic has accused Chinese AI companies DeepSeek and Moonshot AI of secretly routing some customer requests to its Claude models and presenting the resulting answers as outputs from their own systems.
The allegations were included in Anthropic’s latest 154-page threat intelligence report. In a post on X announcing the report, Anthropic called it “our most detailed threat intelligence report to date” and said it had disrupted every operation covered in the document. The company said the cases involved attempted misuse of Claude for cyberattacks, surveillance, influence operations, biological research, weapons development and illicit model distillation.
The latest allegations go beyond previous claims that Chinese AI companies used Claude responses to train competing models through a process known as distillation. Anthropic now says some companies also sent actual customer requests to Claude without those users knowing.
Moonshot Allegedly Sent 300,000 Requests to Claude
Anthropic alleged that Moonshot, developer of the Kimi models, routed almost 300,000 customer requests to Claude over 10 days using 5,380 accounts that Anthropic considered fraudulent.
Most of those accounts appeared to be based in Singapore and Japan, and most requests were directed to Claude’s Opus models.
Anthropic claims Moonshot displayed Claude’s responses to customers as if Kimi had generated them. It also allegedly saved some of those exchanges and used them to extract Claude’s reasoning for training its own models.
The company said the activity exposed sensitive information belonging to Moonshot customers.
One case allegedly involved a user Anthropic assessed as likely affiliated with China’s People’s Liberation Army who believed they were using Kimi to analyse CCTV footage tracking an individual in Chengdu.
Anthropic says it observed more than 23 million exchanges attributable to Moonshot between May and July 2026.
We’re publishing our most detailed threat intelligence report to date.
It covers how people tried to misuse Claude—for cyberattacks, influence operations, surveillance, biology, and building weapons—and how we found and stopped them.
We disrupted every operation in the report,…
— Anthropic (@AnthropicAI) September 10, 2026
DeepSeek Also Accused
Anthropic made similar claims about DeepSeek, saying it silently redirected selected customer requests to Claude without informing those users.
The report says some of the information passed to Claude included internal documents belonging to a Chinese technology company, credentials linked to a Russian government database and data used by engineers working on a Chinese police surveillance system.
Anthropic says it observed more than 12.1 million exchanges attributable to DeepSeek during a 14-day period in July.
DeepSeek and Moonshot had not publicly provided detailed responses to the allegations at the time of reporting.
Xiaomi and Alibaba Also Accused
Anthropic’s report also accused several other Chinese AI companies of large-scale unauthorized distillation, including Alibaba, Z.ai, Xiaomi and MiniMax.
Alibaba allegedly accounted for the largest campaign identified by Anthropic, with more than 151 million exchanges observed between May and July 2026.
Distillation itself is a common AI training technique in which outputs from a more capable model are used to improve another model.
Anthropic describes these particular campaigns as “illicit” because it says they relied on large numbers of fraudulent accounts, proxy services and, in some cases, stolen payment details or API credentials to bypass its restrictions.
US-China AI Dispute Escalates
The report comes shortly after the NSA, FBI and other US agencies accused several Chinese AI companies of conducting industrial-scale efforts to obtain technology and outputs from American AI models. China rejected those allegations and accused Washington of trying to suppress its AI industry.
Anthropic says the alleged rerouting creates another concern beyond intellectual property: users may have unknowingly sent sensitive information to a third-party American AI provider while believing they were communicating only with a Chinese service.
For now, these remain allegations made by Anthropic, and the companies named in the report have not publicly provided detailed rebuttals to each of its claims.
The post DeepSeek and Moonshot Accused of Secretly Routing Over 300,000 Users to Claude appeared first on ProPakistani.
