JPMorgan Chase has stopped its Hong Kong staff from using Anthropic’s AI models, following Goldman Sachs down the same nervous little rabbit hole.
The move shows how tightly banks are watching generative AI outside the US, especially where licensing terms and China-related restrictions get messy.
According to the Financial Times employees at the bank in the Asian financial hub can no longer pick Claude models from an internal “drop-down” list of approved large language models.
One person briefed on the decision said it came from the wording of Anthropic’s usage terms in its licensing deal with JPMorgan.
Blocking access to some of the world’s more advanced AI models could damage Hong Kong’s attempt to rebuild itself as an international financial centre.
That looks particularly awkward as banks elsewhere are using the tools quickly, especially for coding and other software work.
Earlier this year, Goldman Sachs stopped bankers in Hong Kong from using Anthropic’s models, apparently because it took a strict reading of Anthropic’s terms, which exclude use in Greater China, including the former British territory.
Western AI models, including OpenAI’s ChatGPT and Anthropic’s Claude, are blocked in mainland China by the Great Firewall, which keeps outside information on a tight leash. Hong Kong has operated with far less Chinese censorship, but restrictions there are being shaped by US AI companies themselves.
The models cannot be reached directly from Hong Kong, but international outfits have worked around geography by cutting global contracts and hosting activity outside China.
US AI groups are wary of China getting at their models, partly because of “distillation”, where local players train rival systems through intense use of foreign examples.
