Thursday - April 30, 2026
DELFINEO Value Investing Research & News
EN / FR
← Back to news

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

— Summary

Goldman Sachs has stopped its Hong Kong bankers from using Anthropic’s Claude AI models — the latest sign of how the technology is colliding with US-China tensions. Employees in the territory have been unable to access Claude either directly or through Goldman’s in-house AI platform for the past few weeks, according to four sources familiar with the decision.

Western models such as OpenAI’s ChatGPT and Anthropic’s Claude are banned in mainland China by the so-called Great Firewall, but Hong Kong has historically operated outside that perimeter, with usage limits set by the AI companies themselves. One person familiar with the move said Goldman had taken a strict reading of its contract with Anthropic following a consultation with the start-up, concluding that no Goldman employee in Hong Kong should access any Anthropic product. The restriction does not extend to other vendors such as OpenAI. Anthropic told the FT its Claude models had never been officially “supported” in Hong Kong but declined to elaborate. Goldman declined to comment.

The crackdown reflects rising fear of “distillation” — Chinese actors training new models on intensive use of foreign ones. OpenAI accused DeepSeek of doing so last year, and the White House this month accused China of “industrial-scale” theft of US AI labs’ IP — claims Beijing called “pure slander”. The move comes as Anthropic’s new Mythos model raises concern over its potential to crack current cyber defences and threaten financial-system security. Source: Financial Times, 29 April 2026, Arjun Neil Alim, Zijing Wu and Tim Bradshaw.

Why it matters

Hong Kong’s value proposition has long been to operate outside the mainland’s digital perimeter — a place where global banks coordinate cross-border M&A, share sales and trading. If American AI companies and their large customers start treating Hong Kong like the mainland, that proposition erodes. Bankers who use Claude for coding and financial modelling would now lag teams in London or New York.

Distillation fear is the driver: Western firms worry Chinese actors could train rival models by querying foreign systems intensively. OpenAI accused DeepSeek of doing this last year; the White House this month accused China of “industrial-scale” IP theft from US AI labs (Beijing called the claim “pure slander”). Anthropic’s new Mythos model, capable of breaching current cyber defences and stressing financial-system security, has raised the stakes.

Takeaway

A two-tier AI world — open West, restricted China — is hardening, with Hong Kong drifting onto the wrong side of the line for some products. Watch which other banks and corporates follow Goldman’s strict reading; that determines whether this is a Goldman-specific reading or an industry shift.

Further reading

All stories →