Following Goldman Sachs, JPMorgan Blocks Hong Kong Staff from Accessing Anthropic Models
Taylor Wilson
JPMorgan has blocked its Hong Kong employees from using Anthropic's AI model Claude, becoming the second Wall Street bank to shut down AI access in the city. This means → Hong Kong is being steadily locked out of the world's most advanced AI tools.
What exactly did JPMorgan do?
The Financial Times, citing three people familiar with the matter, reported that JPMorgan removed Claude from the internal LLM dropdown available to Hong Kong staff.
The move stems from a strict reading of geographic clauses in JPMorgan's licensing agreement with Anthropic — clauses that explicitly exclude use in Greater China, including Hong Kong.
Goldman Sachs had already taken the same step. This means → two of Wall Street's largest banks read the same contract and reached the same conclusion: Hong Kong is out of scope.
How did Hong Kong access these models before?
In mainland China, Western AI models like ChatGPT and Claude are blocked by the Great Firewall — a system that filters overseas internet content. Hong Kong is technically subject to similar geographic restrictions.
International banks found a workaround: sign a global contract, host actual usage on offshore servers, and sidestep the geographic terms.
In plain terms = the contract was signed in New York, the servers ran overseas, and Hong Kong staff used the tools as if location didn't matter. That channel is now closing.
What does this mean for Hong Kong?
Analysts warn that prolonged restrictions on Hong Kong's access to frontier AI models could erode the city's competitiveness as an international financial hub.
This reflects a widening gap: other global financial centers are rapidly adopting AI-assisted coding and analysis, while Hong Kong is being forced to fall behind.
In plain terms = when bankers in New York and London use AI to write code and run analysis, their Hong Kong counterparts may be stuck with inferior alternatives.
Why are U.S. AI companies so cautious?
U.S. AI companies have long been wary of Chinese access to their models. The core fear is "distillation" — local firms mass-querying a foreign model to generate training data, then using that data to build a domestic competitor.
In plain terms = distillation treats someone else's model as a teacher — ask it thousands of questions, collect the answers, and train your own student to replace it.
Last week Anthropic went further: under a U.S. government directive citing national security, it suspended all access to its most advanced model, Fable. Officials had also raised concerns that Anthropic's Mythos model might be capable of uncovering cybersecurity vulnerabilities in the global financial system.
Content is for reference only, not financial advice.