Goldman Sachs Blocks Claude Access for Hong Kong Staff Amid US-China Tensions

When Goldman Sachs employees in Hong Kong tried to access Claude through the bank's internal AI platform last month, they found it gone. The bank had pulled the connection after consultations with Anthropic, concluding that its Hong Kong staff should not have access to any Anthropic products. The move left the city's Goldman staff in an unusual position: their colleagues in New York, London, and Singapore could still use Claude through the same gateway, while tools like ChatGPT and Google Gemini remained accessible in Hong Kong without restriction. Anthropic had simply drawn a line the bank had not previously noticed.

A Quiet Removal, Then Silence

The removal came without public announcement from either company. Goldman employees in Hong Kong had been accessing Claude through the firm's proprietary AI gateway, an internal hub that routes staff to a curated set of approved external models. Access disappeared a few weeks before the story surfaced in the Financial Times in late April. Goldman offered no comment; Anthropic told the FT that Claude had never been officially "supported" in Hong Kong, while declining to elaborate on its reasoning.

That distinction between "available" and "supported" matters more than it may appear. Anthropic publishes, and updates, a list of the markets where its API and Claude.ai products carry official support, and Hong Kong is absent from it. Whether that absence reflects regulatory caution, legal analysis, business calculus, or all three is something the company has not publicly explained. For Goldman's affected staff, the practical effect was the same regardless: a tool they had been using in daily workflows was gone, with no substitute offered in its place.

A Geopolitical Fault Line Inside an Investment Bank

The timing carries weight. Anthropic spent the first months of 2026 in a public standoff with the United States Department of Defense, which designated the company a national security supply chain risk after Anthropic refused to remove safety guardrails from Claude's military contract. That dispute drew intense scrutiny from Washington and allied governments over which AI systems could be trusted in sensitive national security contexts. It also focused attention on Anthropic's data handling and the geography of its operations.

Hong Kong sits at a particularly sharp intersection of those pressures. Under China's national security law, enacted in 2020, data held or processed by companies operating in the territory is potentially subject to demands from Beijing. For an AI company with strict data-handling policies, a pending IPO in the United States, and a reputation built on safety and trust, extending official support into Hong Kong represents a level of exposure that Anthropic has apparently decided exceeds any commercial benefit. What makes the Goldman case unusual is that the decision surfaced not through a formal policy announcement, but through the bank's own compliance review of its contractual terms with Anthropic.

"Claude models have never been officially supported in Hong Kong." Anthropic spokesperson, as reported by the Financial Times, April 2026

The Map Every Enterprise AI Team Will Now Need to Read

For enterprise clients building global AI workflows around Claude, the Goldman episode is a practical alert about something that has largely stayed off procurement checklists: AI model providers have geographic footprints, and those footprints carry legal and operational implications that not every organization has mapped.

Anthropic is not alone in drawing these lines. OpenAI restricts access across a list of countries tied to US export controls and its own business decisions. Google has its own geographic constraints on Gemini in certain jurisdictions. What the Goldman story makes unusually visible is how those restrictions can emerge quietly inside a large institution's internal tooling, without a formal notification. A product cleared for the New York trading desk may not cover the equivalent team in the regional office a time zone away.

Companies planning cross-border enterprise Claude deployments now have a clear due-diligence item to add: verify geographic coverage explicitly, at the product level and in the contract, rather than assuming a US-listed enterprise agreement covers every office on the org chart. As Claude becomes more deeply embedded in core workflows across banking, law, and professional services, the geographic scope of a vendor's support terms will matter as much as the model's capabilities. The Goldman case was discovered only because the bank ran a thorough compliance review. Other organizations may find the same gap in less controlled circumstances.

Anthropic's posture toward Hong Kong reflects a broader dynamic Dario Amodei has described publicly: US AI leadership carries both commercial opportunity and geopolitical responsibility. Drawing a firm line around Hong Kong is consistent with that view. It is also, for any global enterprise that relies on Claude, a concrete reminder that the line exists.