Detailed Analysis
Goldman Sachs has restricted its Hong Kong-based employees from accessing Anthropic's Claude AI assistant, following a strict interpretation of the bank's existing contract with the AI company. The move came after Goldman Sachs directly consulted with Anthropic and determined that its Hong Kong staff fell outside the permissible scope of Claude usage under the terms of their agreement. Employees who had previously accessed Claude through Goldman's internal AI platform lost that access in recent weeks. Notably, the restriction is specific to Anthropic's products — other AI tools available on the bank's internal platform, including OpenAI's ChatGPT and Google's Gemini, remain accessible to Hong Kong-based staff, underscoring that this is a targeted contractual and geographic limitation rather than a broader pullback from AI adoption.
Anthropic's own characterization of the situation adds a layer of nuance: the company stated that its Claude models had never been formally "supported" in Hong Kong, while declining to elaborate further. This suggests that Hong Kong may have existed in a grey zone within Anthropic's geographic licensing framework — tolerated informally but never officially sanctioned. The bank's direct consultation with Anthropic, which then led to the access removal, implies that Goldman sought explicit clarity and received a definitive answer that triggered the restriction. The episode reveals how enterprise AI contracts, particularly those negotiated at the corporate level, can contain geographic carve-outs or ambiguities that create downstream access disparities across a firm's global workforce.
The development is particularly significant given Hong Kong's unique regulatory position. Unlike mainland China, where US-developed AI tools face strict government-imposed prohibitions, Hong Kong has generally maintained a more permissive environment for access to American AI products. In this case, the restriction originates not from local regulatory mandate but from the private contractual relationship between a US financial institution and a US AI company — a dynamic that reflects the growing complexity of how AI vendors structure their commercial agreements across jurisdictions. The fact that this limitation was self-imposed through contract interpretation, rather than externally mandated, points to how AI companies themselves are increasingly becoming arbiters of geographic access through licensing terms.
More broadly, this incident highlights the emerging tension between the rapid, firm-wide deployment of AI tools at major financial institutions and the fine-grained geographic and regulatory constraints embedded in vendor agreements. As banks like Goldman Sachs accelerate AI integration across global operations, inconsistencies in access — whether driven by contract terms, regulatory exposure, or vendor policy — risk creating operational disparities between regional offices and raising questions about equitable capability distribution within large organizations. For Anthropic, the episode draws attention to the geographic scope of its enterprise agreements at a time when the company is competing aggressively with OpenAI and Google for high-value financial sector clients. How Anthropic structures and communicates its geographic licensing terms going forward could become a meaningful competitive consideration as enterprise clients evaluate AI vendor reliability and transparency across their international footprints.
Read original article →