Goldman Sachs has removed Anthropic’s Claude from its internal AI platform for staff in Hong Kong. The change, reported on April 29, 2026, was not driven by a security breach or a regulatory order. It came down to a strict reading of Goldman’s contractual agreement with Anthropic.

That distinction matters. And if you are running AI tools across your business, it should get your attention.

What Actually Happened

Employees at Goldman Sachs in Hong Kong were previously able to access Claude through the bank’s internal AI platform, where it was primarily used by software engineers for coding. In recent weeks, access was quietly removed.

The reason: after consulting with Anthropic, Goldman concluded that its contract with the company did not support use of Claude products in Hong Kong. Anthropic confirmed that its Claude models had never been officially “supported” in Hong Kong and declined to comment further.

Other AI tools including Gemini and ChatGPT remained available on Goldman’s platform. The restriction applies specifically to Anthropic’s products, and it is location-specific. Staff visiting Hong Kong from overseas could not access Claude while physically in the city.

The Bigger Context: US-China AI Tensions

Goldman’s move did not happen in isolation. Hong Kong sits at the centre of escalating friction between the US and China over AI technology, data security, and access to advanced computing infrastructure.

Banks and financial institutions operating in Hong Kong face increasing scrutiny over where data flows and which technology partners they use. As US-China tensions have deepened through 2025 and 2026, global companies have had to make harder choices about which tools they can deploy in which markets.

For Goldman, a bank with a major Asia Pacific hub in Hong Kong, that created a real operational problem. And it found the answer buried in the fine print of a vendor contract.

Why This Matters Beyond Finance

You might read this and think it is a Goldman Sachs problem, or a financial services problem, or a Hong Kong problem. It is none of those things exclusively.

This is an enterprise AI contract problem, and it is one that most businesses are not paying enough attention to.

When companies adopt AI tools, they focus on capabilities, pricing, and integration. What they often skip is a careful read of geographic restrictions, data residency clauses, and supported regions listed in service agreements. These are often buried in terms of service or supplemental documentation.

For a global bank with an army of lawyers, this still slipped through. For a mid-sized business running AI tools across multiple offices or markets, the risk is proportionally higher.

Here is what the Goldman situation illustrates clearly. AI vendors define where their products are officially supported, and that definition has legal weight. If your team is using a tool in a market where it is not officially supported, you are operating in a contractual grey zone. That creates risk for data compliance, liability, and continuity.

What This Means for Business

Audit your AI tool agreements. If your business operates across multiple locations, pull the terms of service for every AI tool your team uses and check the supported regions list. Pay particular attention to Asia Pacific, where regulatory and geopolitical dynamics are shifting fastest.

Ask about data residency. Where does your data go when it enters an AI tool? Which servers, in which countries, under which jurisdictions? This matters for compliance and increasingly matters for your own contractual obligations to clients.

Understand vendor geography. Not every AI product is available in every market. As US-China tensions continue to shape which technology flows where, expect more situations like this one. The question is whether you find out proactively or reactively.

Plan for tool substitution. Goldman still has Gemini and ChatGPT running in Hong Kong. That redundancy matters. If your entire AI workflow depends on one vendor, a geographic restriction or policy change by that vendor creates operational fragility.

The Quiet Professionalism of Getting This Right

What is notable about Goldman’s situation is how little noise it made. No press release, no public statement. A contractual review, a decision, an access change. That is how mature organisations handle compliance.

The lesson for everyone else is that getting to that point of maturity requires doing the work before the issue surfaces. Auditing AI contracts, understanding geographic and regulatory exposure, and building workflows that do not depend on a single vendor are not glamorous tasks. But they are exactly the kind of foundations that keep operations running when the geopolitical ground shifts.

Enterprise AI is not just a technology choice. It is a compliance surface, a data governance question, and increasingly a geopolitical consideration. The sooner businesses treat it that way, the fewer Goldman-style surprises they will encounter.

---

The practical next step is the free Working With Claude field guide. Thirty-two pages covering the ecosystem, Claude Code, and how to govern a rollout properly. Get your copy.