JPMorgan Chase has restricted employee access in Hong Kong to Anthropic’s artificial intelligence models, reported Financial Times.
According to three people privy to the development, staff at the bank in Hong Kong can no longer select Claude from an internal “drop-down” menu of approved large language models.
Go deeper with GlobalData
Access deeper industry intelligence
Experience unmatched clarity with a single platform that combines unique data, AI, and human expertise.
One person familiar with the change said the move was driven by the language in Anthropic’s usage terms within its licensing arrangement with JPMorgan.
The loss of access to leading AI models could complicate Hong Kong’s efforts to reinforce its role as an international financial centre, as these tools are being adopted quickly elsewhere, especially for coding tasks.
Earlier this year, the FT reported that Goldman had also prevented its Hong Kong bankers from using Anthropic. The bank reportedly applied a strict reading of Anthropic’s terms of use, which bar usage in Greater China, including Hong Kong.
Western AI tools including OpenAI’s ChatGPT and Claude are blocked in mainland China under the Great Firewall, which limits access to information from outside the country.
Hong Kong has for many years operated largely without the same censorship regime, although limits on access in this case have come from US AI providers themselves.
These models are not directly available from Hong Kong, but multinational organisations have been able to work around location-based restrictions through global contracts and by hosting activity outside China, the news publication added.
US AI companies are cautious about their systems being used in China, in part because of the risk of “distillation”, a process in which domestic groups could develop new models through extensive use of foreign systems.
Anthropic did not immediately respond to a request for comment. The company previously told the FT its Claude models had never been officially “supported” in Hong Kong.
