Millions Unknowingly Use American AI Technology Abroad
Many customers around the globe are utilizing AI services developed by American companies, often without realizing it.
A recent report from Anthropic introduces some intriguing insights into the global AI landscape, suggesting that the perceived technological prowess of certain foreign competitors may not be as advanced as they claim.
‘Over 12.1 million exchanges observed.’
The report reveals that several Chinese firms and other international participants appear to depend significantly on AI solutions from American companies.
For instance, various Chinese AI enterprises were found to be sending their customers’ queries to Claude—an AI system from Anthropic—and then presenting those replies as if they originated internally.
Take Moonshot AI, which operates the Kimi chatbot. It has been mentioned that the company was “silently” forwarding client requests to Claude rather than processing them independently.
“Moonshot then presented Claude’s answers to its users, who believed they were interacting with a Kimi model, but were actually receiving responses from Claude,” Anthropic reported.
In just a 10-day span, Moonshot forwarded almost 300,000 customer inquiries to Claude, mostly through the AI program Claude Opus, utilizing over 5,000 proxy accounts primarily based in Singapore and Japan.
Another Chinese firm, DeepSeek, was similarly found to be relaying inquiries to Anthropic’s Claude, even while being a user of its service.
Anthropic indicated that one Chinese firm likely had no idea that its information was being redirected to Claude during its internal document analyses, which contained confidential data.
Furthermore, requests from a Russian government IT operator associated with its Ministry of Defense were routed to Anthropic via DeepSeek, potentially revealing sensitive access details linked to a Russian governmental database.
Additionally, a Public Security Bureau in China utilized DeepSeek to utilize Claude in constructing a case management system. An engineer reportedly developed a tool that analyzed a person’s movements against police records, which involved using citizens’ national ID numbers.
“Over a 14-day period in July 2026, more than 12.1 million exchanges attributed to DeepSeek were noted,” Anthropic stated.
Moreover, Xiaomi, a Chinese tech firm, allegedly engaged in “illicit distillation,” making 400,000 exchanges with Anthropic over 20 days to extract data to enhance its AI models.
RELATED: Concerns Rise Over Hackers Targeting X Users
Anthropic also indicated that three accounts linked to the Iranian government were using Claude to orchestrate campaigns aimed at influencing public opinion both domestically and internationally, representing the Islamic Culture and Communications Organization under the Ministry of Culture and Islamic Guidance.
This effort was part of a broader nine-part strategy focused on “international influence,” which included a detailed organizational plan for commemorating Iran’s supreme leader.
