JPMorgan has removed Anthropic’s Claude from the list of approved large language models for employees based in Hong Kong, the Financial Times reports — a move driven not by performance but by licensing language that limits where the models can be used.
What happened
- JPMorgan Chase staff in Hong Kong can no longer pick Claude from the bank’s internal LLM catalogue. People familiar with the decision told the FT the restriction stems from Anthropic’s licensing terms, which define geographic limits on model use. JPMorgan declined to comment.
- The move echoes an earlier decision by Goldman Sachs, which blocked Hong Kong bankers after concluding Anthropic’s terms exclude usage across Greater China (including Hong Kong). Anthropic has said Claude was never officially supported in Hong Kong.
Why it matters to finance and crypto
- Western AI firms generally block or limit access to their most advanced models in mainland China; Hong Kong has historically been looser, letting international firms access frontier models via enterprise agreements and offshore infrastructure. Limits from major banks renew concerns that Hong Kong could lose an edge as AI tools become central to software development, research, trading, and smart-contract and security audits in fintech and crypto.
- Regulators and industry observers are also worried about “model distillation” — foreign actors using access to accelerate domestic AI development — which helps explain tighter geographic controls.
Broader Anthropic context
- Less than a week before the JPMorgan restriction, Anthropic suspended access to its new Fable 5 and Mythos 5 models after receiving a U.S. government export-control directive on June 13. Authorities reportedly ordered the company to block those models for all foreign nationals, including foreign-national employees inside the U.S., amid concerns about a potential jailbreak technique that could let the models identify or repair software vulnerabilities. Anthropic pushed back, calling the significance of the issue overstated and hinting at a possible misunderstanding.
- Two days later a proposed class-action suit was filed in U.S. federal court by plaintiff Karl Kahn, alleging subscribers to Anthropic’s premium Claude plans (Max 5x and Max 20x) paid for promised multipliers of usage that were not delivered. The suit seeks class status for customers who subscribed since April 2024.
Policy posture
- These legal and access issues come as Anthropic itself has publicly called for stronger oversight. In its June 11 “Policy on the AI Exponential,” the company urged governments to require testing, independent evaluation, cybersecurity standards, and enforcement for frontier AI systems, citing biological, cybersecurity, and operational risks.
Bottom line
- Geographic and export constraints on advanced LLMs are tightening, driven by licensing language, national security concerns, and regulatory scrutiny. For finance and crypto firms that rely on LLMs for code review, research, trading models, or smart-contract audits, these developments increase operational uncertainty and could reshape where and how frontier AI is used.
Read more AI-generated news on: undefined/news
