JPMorgan and Goldman Sachs restrict Anthropic Claude access abroad

Brief

The most practical enterprise AI story for June 19, 2026 is JPMorgan and Goldman Sachs restricting Anthropic Claude access for employees outside the United States. JPMorgan has blocked Claude access for staff in Hong Kong, while Goldman Sachs had already taken a similar approach for bankers in the region.

For people comparing AI tools, this matters because enterprise AI compliance is no longer only about data privacy and internal approval. It is now tied to geography, contracts, export controls, model availability, and whether a tool can be used consistently across global offices.

What happened today

JPMorgan Chase and Goldman Sachs have told employees outside the U.S. not to use Anthropic tools. The most visible case is Hong Kong, where JPMorgan staff can no longer select Claude models from the bank's approved internal large language model list.

The decision appears connected to Anthropic usage terms, regional restrictions, and a stricter reading of rules around Greater China. It also lands during a broader U.S. policy fight over advanced AI model access. U.S. officials have pushed for tighter controls on advanced model exports to certain foreign users because of national security concerns.

The result is a new kind of enterprise AI access problem. A tool can be approved in one country and blocked in another, even inside the same global company.

Why it matters

• JPMorgan and Goldman Sachs are treating Anthropic Claude access as an enterprise AI compliance issue.
• Hong Kong is becoming a test case for how global banks handle frontier AI tools under regional restrictions.
• AI export controls are moving from government policy into corporate software access lists.
• Frontier model access now depends on geography, licensing terms, internal risk review, and national security pressure.
• Global companies may need separate AI tool policies for different offices, regions, and user groups.
• Vendors that sell AI tools to enterprises will need clearer language around allowed regions, model availability, and usage restrictions.

What changes for AI tools

This news makes enterprise AI procurement more complicated. Teams can no longer ask only whether a model is powerful, private, or affordable. They also need to ask whether it can be used by every office that needs it.

For AI tool builders, the practical lesson is to make access rules explicit. Enterprise buyers need documentation on geography, model availability, data residency, export controls, audit logs, and what happens when a region becomes restricted.

For buyers, the safe workflow is to build an AI tool register by office and use case. A tool that is allowed for U.S. analysts may not be allowed for Hong Kong employees. A model that works for marketing copy may not be approved for sensitive financial work.

What builders should watch

Builders should watch whether other banks, consulting firms, law firms, and multinationals copy this approach. Finance often sets the tone for regulated enterprise AI usage because banks have strong compliance teams and global exposure.

They should also watch whether Anthropic, OpenAI, Google, and other model providers publish clearer enterprise access matrices. If customers cannot understand where a model is allowed, procurement friction will increase.

What users should watch

Users inside global organizations should not assume that access in one location means access everywhere. If a workflow depends on Claude, ask whether the same model is approved in every region that uses the workflow.

Users should also keep prompts and workflows portable. If one model becomes unavailable in a region, teams should be able to move the work to another approved model without losing process knowledge.

Search intent breakdown

People searching for JPMorgan Goldman Sachs Anthropic Claude today are likely asking why the banks restricted Claude access, whether Hong Kong is affected, and what it means for enterprise AI adoption.

People searching for AI export controls are asking whether government restrictions are changing which AI tools companies can use.

People searching for enterprise AI compliance are asking the Goodiebase question: how do teams choose AI tools when access depends on policy as much as capability?

Goodiebase view

This is practical AI tools news because access is now part of product quality. A model that cannot be used across the offices that need it is not just a compliance issue. It is a workflow reliability issue.

For Goodiebase users, the takeaway is simple: compare AI tools by capability, governance, and regional availability. For enterprise teams, the best AI product is the one that can be used legally, consistently, and transparently where the work happens.