Anthropic Struggles to Block Unauthorized Access by Chinese Firms

▲ Anthropic

As Chinese companies increasingly use various methods to gain unauthorized access to advanced U.S. artificial intelligence (AI) models, Anthropic has significantly strengthened its measures to block them.

The Financial Times reported that despite strict U.S. export controls and service restrictions, Chinese engineers are using various methods to access the latest U.S. AI technology. Anthropic is reportedly identifying and blocking the actual methods used by Chinese users, such as shutting down proxy connections or intermediary service accounts.

Recently, it was reported that the Chinese fintech giant Ant Group created corporate accounts for Anthropic's Claude using the names of entities established overseas, such as in Singapore, to allow employees at its Chinese headquarters to access the service.

Ant Group enabled its employees to access corporate Claude accounts through an internal intranet connected to its Singaporean subsidiary.

Chinese companies are also known to have accessed AI tools such as Anthropic's Claude through cloud services, including Microsoft Azure.

Microsoft sold application programming interface (API) access to Chinese companies based in Singapore, which allowed engineers at their Chinese headquarters to use Claude via internal networks.

One source told the newspaper, "Accessing Claude through overseas subsidiaries is a known issue and is not limited to any specific company."

While such workaround connections do not violate U.S. or Chinese laws, they do breach Anthropic's terms of service, which prohibit the use of its models by Chinese companies and their overseas subsidiaries.

ByteDance, the parent company of TikTok, did not provide direct access to Claude, but it allowed engineers to expense the costs if they subscribed to the service individually using a Virtual Private Network (VPN).

In effect, the company supported employees working in China to disguise their internet protocol (IP) addresses as being in countries where Claude services are permitted, such as the U.S. or Singapore.

Anthropic stated that it is taking a firm stance, noting that it continuously updates its tools to detect and sanction new workaround techniques as Chinese users develop them.

Anthropic monitors detailed metrics, such as the time zone of the computer used for access, and analyzes the country of issuance for credit cards used for payment to identify and block fake accounts that appear to be in the U.S. by IP address but are actually accessing from China.

"We explicitly prohibit accessing or facilitating access to Claude in regions where it is not supported, including China," Anthropic stated. "We are the only leading AI company that restricts services to companies owned by Chinese firms, even those established outside of China."

The company added, "We enforce this policy through our continuously evolving detection systems and work with our partners to identify and block accounts that violate our policies."

The Chinese government also prohibits domestic companies from using models located in overseas data centers to develop consumer applications for security reasons, but it does not prevent Chinese AI research institutes from using foreign models for internal research and development purposes.

(Photo: AP, Yonhap News)

※