Goldman Sachs has removed access to Anthropic’s AI models for its banking staff in Hong Kong, reported Reuters citing a source with direct knowledge of the matter.
The source said employees in Hong Kong had previously been able to use Anthropic’s Claude model through an internal AI platform, but that access has been unavailable in recent weeks.
The Financial Times first reported the change, citing people familiar with the situation.
According to the FT, Goldman took the step after adopting a strict reading of its contract with Anthropic following consultations with the AI company.
The newspaper said the bank concluded its Hong Kong employees should not be permitted to use Anthropic products.
The Reuters source said other widely used AI models, including Gemini and ChatGPT, remained accessible on the bank’s internal platform.
Anthropic’s latest AI model, Mythos, has attracted attention from global banks and financial regulators because of concerns about possible risks to banking systems.
Reuters reported last week that the Hong Kong Monetary Authority said some lenders were considering extra safeguards in response to changing cyber security threats linked to advanced AI models.
Reuters said it could not verify whether other banks or companies had also restricted access in Hong Kong.
Hong Kong does not currently appear on Anthropic’s list of markets where its API and Claude.ai are officially available, according to the company’s disclosures.
The move comes against a backdrop of growing friction between the US and China over AI technology, data security and access to high-end computing tools.
Although AI models such as ChatGPT and Claude, developed by US companies, are banned in mainland China, Hong Kong has largely remained outside those restrictions, with access limits determined by the US companies themselves.
An Anthropic spokesperson told the FT that its Claude models had never been officially “supported” in Hong Kong, while declining to comment further.
Several major AI groups have raised concerns that their models could be used by Chinese competitors for training purposes.
In 2024, OpenAI restricted traffic from China to its application programming interface, which developers use to incorporate its AI models into other products.
