Wall Street's AI ambitions just hit a geographic wall. Goldman Sachs has blocked its Hong Kong staff from using Anthropic's Claude after determining the bank's contract with the AI company doesn't cover the Chinese territory, Bloomberg reported. The restriction is location-specific. Staff visiting Hong Kong from overseas also lose access to Claude while in the city, according to a person familiar with the matter. The technology was predominantly used by software engineers, the person said.
Goldman arrived at the decision after consulting with Anthropic and taking a strict interpretation of their contractual arrangement, the Financial Times reported. The bank concluded its employees in Hong Kong, its main Asia Pacific business hub, should not use any Anthropic products.
Other AI models remain untouched. Google's Gemini and OpenAI's ChatGPT are still accessible on Goldman's internal platform, a source told Reuters.
Anthropic's position adds context. A company spokesperson told the FT that Claude models had never been formally "supported" in Hong Kong, but declined to elaborate further. The move is not an AI retreat. Goldman has spent the past six months embedding Anthropic engineers within the firm to co-develop autonomous agents for accounting, trade transactions, and client vetting, according to CIO Marco Argenti.
He described the initiative as creating "a digital co-worker for many of the professions within the firm that are scaled, are complex, and very process-intensive."
CEO David Solomon has made AI central to the bank's strategy, and the Hong Kong restriction does not affect Goldman's broader Anthropic partnership. But it exposes a compliance gap that other multinational banks with Anthropic enterprise agreements in Hong Kong may now need to audit: their AI contracts, written before US-China tech tensions escalated, never accounted for geography as a compliance variable.
Mainland China already bans US-developed AI models like ChatGPT and Claude under the Great Firewall. Hong Kong has historically operated outside those controls. That line is blurring, and enterprise AI contracts are catching up to the reality.
