Major US financial institutions operating in Hong Kong are aggressively rolling out advanced artificial intelligence tools to their employees, navigating a complex geopolitical landscape where top American AI firms have restricted access to the region. According to SCMP, despite companies like OpenAI and Anthropic officially barring users in Hong Kong and mainland China from accessing their models directly, banks such as Citigroup and Morgan Stanley are deploying proprietary platforms powered by these very technologies to maintain a competitive edge.
Citigroup’s “Arc” Agent Platform
Citigroup Hong Kong recently began trialing a new in-house AI agent platform called “Arc,” initially targeting local software developers and engineers. Arc represents a significant leap forward from previous tools, enabling the end-to-end automation of complex tasks such as client prospecting, portfolio data gathering, and scenario modeling. Senior management has been described as “aggressively” encouraging AI adoption since last summer. The bank’s broader proprietary platform, “Citi AI,” utilizes both Google Gemini and Anthropic’s Claude. To bypass regional restrictions, Citigroup reportedly routes its Hong Kong employees through overseas servers, such as those in Japan, to access Claude. One Citigroup employee described the impact of AI agents: “AI agents are the new thing that really makes me feel like everything could change. Previously, I would have to prompt the AI through multiple turns, but now it looks like the AI agents can do this on their own.”
Morgan Stanley’s Multi-Model Approach
Morgan Stanley continues to offer its Hong Kong employees access to Claude and OpenAI’s GPT models across several divisions. SCMP reported that at an internal “global technology expo” held at its Asia-Pacific headquarters in Hong Kong in late April, the bank showcased over a dozen AI innovations, demonstrating tools powered by Claude, GPT, xAI’s Grok, and Microsoft’s Copilot. One notable application is “CyberMind,” which leverages AI model reasoning to orchestrate agents for cybersecurity troubleshooting. This multi-model strategy reflects a broader industry trend toward leveraging the best available AI tool for each specific task, rather than committing exclusively to a single vendor.
The aggressive adoption by US banks contrasts sharply with the cautious approach of mainland Chinese lenders operating in Hong Kong, who remain wary of internal AI usage due to stringent data compliance concerns, limiting their tools to basic chatbots. The deployment of these advanced tools also remains strictly limited to Hong Kong. Citigroup has not rolled out any in-house AI tools to its mainland employees, causing frustration among staff who lack access to generative AI for routine tasks. This disparity reflects the broader challenge facing multinational corporations operating across the US-China divide, where the travel restrictions on AI talent and access restrictions on AI tools are creating an increasingly fragmented global technology landscape.
The aggressive AI deployment by US banks in Hong Kong also raises important questions about regulatory oversight and data governance. Hong Kong’s financial regulators, including the Hong Kong Monetary Authority (HKMA), have been developing guidelines for the responsible use of AI in financial services, but the pace of adoption by major institutions is outstripping the development of formal regulatory frameworks. The use of overseas servers to circumvent regional access restrictions, while technically compliant with current rules, underscores the need for clearer guidance on data sovereignty, model accountability, and the governance of AI-assisted financial advice. As AI becomes more deeply embedded in the operations of global financial institutions, the regulatory and geopolitical dimensions of its deployment will become increasingly difficult to separate. Hong Kong’s unique position as both a global financial center and a Special Administrative Region of China makes it the world’s most consequential laboratory for this challenge, and the choices made by regulators and institutions there will set precedents that reverberate across the global financial system.
The contrast between the aggressive AI adoption of US banks and the cautious approach of their Chinese counterparts in Hong Kong also illuminates a broader competitive dynamic. In the short term, the banks with the most capable AI tools will be able to serve clients more efficiently, generate more accurate insights, and execute transactions faster, creating a performance gap that will be difficult for less AI-enabled competitors to close. In the longer term, the institutions that build the most sophisticated AI-augmented workflows today will accumulate the institutional knowledge, proprietary data, and organizational capabilities needed to leverage the next generation of AI tools when they arrive.
Goldman Sachs’s decision to bar its Hong Kong bankers from using Claude, reportedly due to a strict reading of its vendor contract, illustrates the risk of allowing compliance caution to become a competitive liability. In the race to deploy AI in financial services, the cost of moving too slowly may ultimately prove higher than the cost of moving too fast. Hong Kong’s financial sector is, in this sense, a microcosm of the broader global challenge: navigating the intersection of technological opportunity, geopolitical constraint, and regulatory uncertainty in real time, with no clear playbook to follow.