Goldman Sachs and OKX have cut Hong Kong-based staff off from Anthropic's Claude AI model, each citing the company's geographic access rules that keep the tool out of Hong Kong and mainland China. The restrictions show how a US AI company's compliance stance is colliding with the financial industry's rush to use the technology.
Why the ban hit
Anthropic doesn't allow Claude to be used in Hong Kong or mainland China. That's the root of the problem for both firms. OKX barred employees based in Hong Kong or traveling through China from using Claude, after briefly suspending its corporate Anthropic account. The account is back, but the access limits remain. Goldman cut Claude access for Hong Kong staff earlier in 2026, concluding its Anthropic contract didn't cover the territory. The restriction also applies to overseas staff when they visit Hong Kong.
OKX's workaround
OKX isn't going to help staff sneak around the rules. The exchange said it won't assist employees in using VPNs to reach Claude, and it will route their AI requests to other models instead. That's a notable stance for a firm that has pushed AI use hard. OKX has tied AI usage to performance reviews, and its CEO, Star Xu, has said the company spends $6 million to $8 million a month on large language model providers. The mandate isn't unique to OKX; BeInCrypto reported that rival exchanges also require engineering teams to use AI tools.
Goldman's contract snag
Goldman's problem is contractual, not a suspension. The bank determined its Anthropic agreement didn't cover Hong Kong, so it cut access there. That doesn't mean Goldman is backing away from Claude. Its CIO, Marco Argenti, has embedded Anthropic engineers to build AI agents for trade accounting and client vetting. Software engineers made up the bulk of Claude's users at Goldman, so the restriction hits a specific group.
Wider friction
The Hong Kong government has been encouraging financial firms to adopt AI, but these geographic restrictions complicate that push. Neither company has said whether the limits extend to mainland China-based clients or partners. US and Chinese officials are slated to hold AI talks in September, which could clarify how these tools are governed across borders.
