JPMorgan blocks Hong Kong staff from accessing Anthropic AI models

JPMorgan Chase stopped staff in Hong Kong from accessing Anthropic’s AI models, including Claude, after the wording of Anthropic’s usage and licensing terms in the bank’s agreement prompted a review. The models were removed from an internal drop-down list of approved large language models available to employees in Hong Kong, a sign that access to frontier AI is becoming a controlled compliance decision inside one of the world’s largest banks.

The move matters beyond one vendor or one office. Global banks are increasingly using generative AI for research, coding, document analysis and customer support, but those same tools raise immediate questions about data security, client confidentiality and regulatory expectations. Hong Kong sits at a particularly sensitive intersection of those pressures, with major international banks running regional teams there while managing Chinese market exposure, local privacy rules and cross-border compliance demands.

That is why the decision looks less like a retreat from AI than a tighter sorting process, with banks approving some tools in some markets and restricting them in others. Hong Kong regulators have already signaled that AI use in finance will not be treated as a free-for-all. The Office of the Privacy Commissioner for Personal Data has published AI security resources and guidance to help organizations write internal policies for generative AI in line with the Personal Data Ordinance. The Hong Kong Monetary Authority has also issued consumer-protection guidance on the use of generative AI by authorized institutions.

JPMorgan’s step follows a similar move by Goldman Sachs, which removed access to Anthropic’s Claude for bankers in Hong Kong in April 2026 as scrutiny intensified over data security and cyber risks. Taken together, the decisions point to a broader shift in the banking sector, where access to the same AI model can depend on where an employee sits, what a contract allows and how regulators in that jurisdiction interpret the risks.

For Anthropic, the episode shows that enterprise adoption is no longer determined only by model performance. Usage terms, licensing language and regional risk tolerances are becoming just as important. For global finance, the result is a fragmented AI landscape shaped by U.S.-China tensions, local regulation and the growing view that frontier AI is not a universal utility, but a controlled asset with borders.