Goldman Sachs traders in Hong Kong opened the bank's internal AI platform on Monday morning and found that Claude had disappeared.
Gemini was still available. ChatGPT was still available. But Anthropic's Claude, the model many teams used for coding work and investment-banking workflows, had quietly been removed from the access menu. Reuters and the Financial Times both reported the move on April 29, framing it less as a technical outage than as a contractual boundary.
Not a technical failure, but a contract issue
The cutoff was not an Anthropic ban, and it was not Goldman abandoning Claude globally. Goldman reviewed its agreement with Anthropic and confirmed a simple point with Anthropic's legal team:
Hong Kong is not on the official list of supported regions for Anthropic API access or Claude.ai.
Anthropic's public availability list includes markets such as Japan, Singapore and South Korea, but not Hong Kong. Goldman's lawyers treated that omission literally. If Hong Kong was not covered by the contract, staff in the Hong Kong office could not use the service.
The result is subtle but important. Anthropic did not publicly say Hong Kong users were barred, and Goldman did not say it had lost confidence in Claude. Both sides stayed inside the language of contract compliance, yet several thousand bankers in Hong Kong lost access to one of the bank's most useful AI tools.
Goldman had invested heavily in Claude
| Time | Event |
|---|---|
| February 2026 | Goldman CIO Marco Argenti said the bank was working with Anthropic on AI agents to automate more business functions. |
| Earlier | Claude Code had already been running inside Goldman's global software production pipeline. |
| April 2026 | The Hong Kong office lost Claude access. |
In plain terms, Goldman was a serious Claude customer, with senior executives publicly backing the relationship. The access cut applies only to Hong Kong; other Goldman offices continued using Claude.
Why Hong Kong, and why now
The timing is hard to ignore. A Trump-Xi summit was scheduled for mid-May. In the weeks before it, the Hong Kong Monetary Authority had contacted several major banks to ask whether they had assessed the risks that Anthropic's latest Mythos model could pose to Hong Kong's banking system.
In financial circles, that kind of question is understood as a regulatory signal. By removing access under the contract terms, Goldman effectively cleaned up its compliance position before the issue could become more sensitive.
Banks rarely gamble on cross-border data and regulatory risk. If Hong Kong is absent from Anthropic's service terms, legal teams treat that absence as decisive. When legal says the tool cannot be used, business teams stop using it, even if that means losing Claude Code, Mythos vulnerability research and other capabilities.
No one wants to say it directly
- Goldman: declined to comment.
- HKMA: declined to comment.
- Anthropic: told the Financial Times that Claude models had never been officially supported in Hong Kong.
All three responses avoided the phrase "geopolitics." The market understood the implication anyway.
The problem is larger than one office
Claude's presence inside major investment banks has grown quickly over the past half year. Morgan Stanley, JPMorgan and Goldman have all been rolling Claude Code into internal workflows. Once regional access cuts become a precedent, other sensitive markets, from Singapore to Dubai, may have to revisit their own contracts.
For Anthropic, the trade-off is difficult. Expanding coverage means more compliance, data-residency and legal work across jurisdictions. Keeping coverage narrow gives enterprise customers' legal teams more reasons to remove access one market at a time.
The next test may arrive after the May summit.
Sources: Goldman Cuts Access to Anthropic's Claude for Hong Kong Bankers, CocoLoop; Source Says (Reuters / Yahoo Finance); Goldman Staff in Hong Kong Lose Access to Anthropic's Claude (Bloomberg); Why Goldman Sachs Blocked Hong Kong Bankers from Anthropic's Claude AI (Disruption Banking)