Goldman Sachs stops bankers using Anthropic’s Claude in Hong Kong

Stay informed with free updates

Simply sign up to the Artificial intelligence myFT Digest — delivered directly to your inbox.

Goldman Sachs has stopped its bankers in Hong Kong from using Anthropic’s AI models, in the latest sign of how the emerging technology is brushing up against US-China tensions.

Employees of the Wall Street bank in the Chinese territory were unable to access Claude models either directly via the in-house artificial intelligence platforms as of a few weeks ago, according to four sources familiar with the situation.

Western AI models such as OpenAI’s ChatGPT and Claude are banned in mainland China as part of the so-called Great Firewall. But Hong Kong has long operated mostly outside of Chinese censors and restrictions on usage are imposed by the US AI companies themselves.

One person familiar with Goldman’s move said it came as a result of the US bank taking a strict interpretation of its contract with Anthropic following a consultation with the Silicon Valley start-up. 

That reading concluded that Goldman employees in Hong Kong should not be able to use any Anthropic products. The person said this did not extend to contracts with other AI vendors such as OpenAI.

A spokesperson for Anthropic said its Claude models had never been officially “supported” in Hong Kong but declined to comment further. Goldman declined to comment.

American AI companies are wary of usage of their models in China in part due to the threat of “distillation” in which local actors could train new models through intensive usage of foreign ones. 

OpenAI last year accused Chinese rival DeepSeek of using its models to train its own model, while the White House this month accused China of undertaking “industrial-scale” theft of US AI labs’ intellectual property.

No evidence has been disclosed to support OpenAI’s claim, while the Chinese embassy in Washington said the White House accusations were “pure slander”.

The new curb on Goldman bankers’ usage of Claude could represent a challenge for Hong Kong as a revived financial and knowledge hub if employees, especially those who use Claude for coding and financial modelling, are unable to access the most advanced models and risk falling behind other teams or organisations.

It also poses questions for other companies and institutions in the former British territory that have enterprise deals with Anthropic globally and continue to use its models in Hong Kong. The FT could not confirm whether other banks or companies have also decided to restrict access.

Hong Kong remains the hub for investment banking and finance across Greater China for most global banks, which use the territory as a place to co-ordinate cross-border activity including trading, M&A and share sales.

The crackdown on Anthropic models also comes as the start-up’s new Mythos AI model has raised concern among governments and companies worldwide that it could crack current cyber security systems and pose risks to the global financial system.

With additional contributions by Cheng Leng in Beijing