JPMorgan Chase has removed Anthropic’s Claude models from the suite of large language tools available to staff in Hong Kong, a targeted move that signals how fast AI governance concerns are reshaping financial sector technology choices in Asia’s premier banking hub. Rather than a wholesale retreat from AI, the decision illustrates the growing weight of contractual language, cross border regulation, and geopolitical risk in banks’ selection of specific AI vendors.
A Narrow Block with Broad Signals
According to reports citing people familiar with the matter, JPMorgan employees in Hong Kong can no longer choose Anthropic’s Claude models from the internal list of approved large language models. The trigger was reportedly the wording of Anthropic’s usage terms within its licensing agreement with the bank, which raised enough concern for the institution to delist those models locally.
The move follows a similar step by Goldman Sachs, which recently removed Claude from the set of tools accessible to its Hong Kong bankers, suggesting that this is not an isolated compliance twitch but the start of a pattern among major US institutions in the territory. For now, the restriction is geographically specific, which indicates that the perceived risk is tied to Hong Kong’s regulatory and geopolitical environment rather than to Anthropic’s technology per se.
Contractual Fine Print Meets Geopolitics
That a single clause in a licensing agreement can effectively switch off a widely used model inside part of a global bank shows how AI adoption now hinges on legal nuance as much as technical performance. Usage terms that might be acceptable in New York can become problematic in Hong Kong, where questions around data routing, model training, and third country access are politically charged.
This is unfolding amid heightened tension between the United States and China over advanced AI, data security, and high end computing power, which has already led Washington to restrict exports of powerful models and chips. Recent US government pressure on Anthropic to curb access to some of its most capable models for foreign users on national security grounds shows that vendors serving global clients can be pulled in opposite directions by their biggest regulator and their most demanding customers.
Hong Kong’s AI Push Confronts Compliance Headwinds
JPMorgan’s decision also sits uneasily alongside Hong Kong’s ambition to build itself into a regional hub for responsible AI in finance. The Hong Kong Monetary Authority and other regulators have rolled out a Generative AI Sandbox that offers banks supervisory guidance and access to computing resources to pilot AI use cases in areas such as risk management, anti-fraud controls, and customer engagement.
Regulators have issued a policy statement on responsible AI in financial markets and guidelines stressing that boards and senior managers remain accountable for AI driven decisions, including requirements around data protection, bias mitigation, and explainability. These initiatives aim to encourage experimentation, but they also raise the bar on governance and may make global banks especially sensitive to any ambiguity in vendor terms that could expose them to cross jurisdictional scrutiny.
Strategic Implications for Banks and Vendors
For large institutions, the most immediate implication is that AI strategy is fragmenting by jurisdiction, with different model lineups and risk tolerances in Hong Kong, mainland China, Europe, and the United States. That fragmentation adds complexity and cost: compliance, training, and workflow design all need to adapt to a patchwork of approved tools while maintaining consistent service quality and risk controls.
For AI vendors, the episode underscores that technical capability is not enough; contracts, data residency assurances, and geopolitical exposure now form part of the competitive landscape. Providers that can structure terms to satisfy both US security expectations and Asian financial regulations without over claiming rights to data or model usage will be better placed to survive the next wave of risk reviews by large banks.
A Final Note
JPMorgan’s removal of Anthropic tools in Hong Kong is a modest operational change with outsized symbolic weight, highlighting how quickly AI access can shift when legal phrasing collides with geopolitical caution and evolving local rules. As Hong Kong seeks to champion responsible AI innovation while remaining enmeshed in US China strategic competition, banks and AI firms alike will need to treat governance design and contract language as core elements of product strategy, not afterthoughts.
