JPMorgan restricts Anthropic Claude access for employees in Hong Kong

Anthropic’s AI models have lost another major banking user group in Hong Kong after JPMorgan restricted employee access to Claude under the company’s licensing terms.

The Financial Times reported that JPMorgan Chase employees in Hong Kong can no longer select Anthropic’s Claude models from the bank’s internal list of approved large language models. 

Three people familiar with the matter told the publication that the restriction stems from language in Anthropic’s licensing agreement. One person familiar with the decision said JPMorgan based the move on terms governing where the models can be used.

The development follows a similar decision by Goldman Sachs earlier this year. The Financial Times previously reported that Goldman blocked bankers in Hong Kong from using Anthropic models after determining that Anthropic’s terms of service exclude usage across Greater China, including Hong Kong.

Anthropic has not issued an official statement, but the company has previously told The Financial Times that Claude had never been officially supported in Hong Kong. JPMorgan declined to comment.

Hong Kong access faces new constraints

Western AI companies have generally restricted direct access to their most advanced models in mainland China. OpenAI’s ChatGPT and Anthropic’s Claude are unavailable there because of a combination of company policies and China’s internet controls.

Hong Kong has historically operated with fewer internet restrictions than mainland China. International firms have often obtained access to frontier AI models through global enterprise agreements and infrastructure hosted outside China.

The Financial Times reported that access limitations at major financial institutions have renewed concerns about Hong Kong’s ability to remain competitive as AI tools become more deeply integrated into software development, research, and financial services workflows.

Anthropic’s approach to geographic restrictions comes as U.S. AI companies face increasing scrutiny over how advanced models are used outside the United States. Industry observers and policymakers have expressed concerns that foreign users could employ frontier systems to accelerate domestic AI development through a process commonly known as model distillation.

Anthropic navigates multiple challenges

The banking restrictions arrive less than a week after Anthropic suspended access to its newly released Fable 5 and Mythos 5 models.

Anthropic announced on June 13 that it had disabled both systems after receiving a U.S. government export-control directive. The company said authorities instructed it to block access to the models for all foreign nationals, including foreign-national employees working within the United States.

Anthropic stated at the time that officials were concerned about a potential jailbreak technique that could allow the models to identify or repair software vulnerabilities. The company disputed the significance of the reported issue and said it believed the government action may have resulted from a misunderstanding.

Only two days later, Anthropic became the target of a proposed class-action lawsuit filed in the U.S. District Court for the Northern District of California. The complaint alleges that subscribers to the company’s $100-per-month Max 5x and $200-per-month Max 20x Claude plans received substantially less usage than marketing materials led customers to expect.

Plaintiff Karl Kahn seeks class-action status on behalf of customers who paid for Anthropic’s premium Claude subscriptions since April 2024. The filing argues that usage limits imposed on subscribers did not match the multipliers promoted for the plans.

Those disputes emerged shortly after Anthropic publicly called for stronger regulation of frontier AI systems. In its June 11 “Policy on the AI Exponential” proposal, the company urged governments to establish testing requirements, independent evaluations, cybersecurity standards, and enforcement mechanisms for the most advanced AI models.

Anthropic argued in that proposal that frontier systems can introduce biological, cybersecurity, and operational risks that require closer oversight as AI capabilities continue to advance.