US Banks Push American AI into Hong Kong Despite Geopolitical Crossfire

Arc is designed to allow the bank's developers to build and scale autonomous AI agents for specific internal use cases, with an initial focus on automating manual, complex tasks such as research, data synthesis, market analysis, and client-preparation work . This includes the kind of end-to-end workflow automation for client prospecting that is highly valuable in the competitive Hong Kong market. The platform acts as a secure, walled garden where multiple agents can collaborate on tasks, with managers able to monitor behavior and intervene to manage risk—a critical feature given the sensitivity of financial data .

"For the first time, we can deploy embedded AI agents at enterprise scale across every business line, every geography, every function," said Citi's CTO, David Griffiths, underscoring the platform's global ambition that directly includes its Hong Kong operations . Over 80% of Citi's global staff were already regularly using in-house AI tools before Arc's launch, signaling that the bank is not experimenting at the edges but embedding AI deeply across its operations .

The Compliance Contradiction: Goldman Sachs Restricts Claude

In stark contrast to Citi's full-speed-ahead proprietary build, Goldman Sachs demonstrated the other side of the geopolitical calculus in late April 2026. The bank restricted its Hong Kong-based bankers from using Anthropic's Claude AI, following an internal contract review . The move was not a blanket ban on AI; staff were still permitted to use ChatGPT and Gemini for internal workflows. It was a targeted restriction driven by a specific risk assessment .

This decision unfolded against a backdrop of rising AI tensions, where US AI companies are increasingly accusing Chinese firms of using their models to train competing systems . The move by Goldman Sachs, as noted by Semafor, suggests that US AI firms and their corporate clients are starting to view Hong Kong not as a separate entity, but as an extension of mainland China's regulatory sphere when it comes to AI model access . This action creates a compliance contradiction: banks are pushing ahead with AI adoption while simultaneously walling off parts of their toolkit, forcing them to become expert geopolitical risk managers.

The Broader Landscape: Productivity Amid Uncertainty

The strategic whiplash between Citi's buildout and Goldman's pullback is happening against an uncertain economic backdrop. KPMG's Hong Kong Banking Outlook 2026 notes that while the city's banks enter the year from a position of strength, their key challenge is how to deploy capital amid ongoing US-China tensions, tariffs, and diverging monetary policies .

The report forecasts that the AI focus in 2026 will be less on innovation for its own sake and more on practical steps: measuring ROI, delivering productivity gains, and embedding AI "front-to-back" across operations . This pragmatic demand for secure, measurable returns makes the case for a proprietary platform like Citi's Arc even stronger. It offers controlled productivity gains without the sudden geopolitical whiplash that can come with a third-party tool restriction.

The New Calculus: Build In-House, Restrict Selectively

The actions of Citigroup and Goldman Sachs are not contradictory; they are complementary elements of a single, emerging strategy for US banks in a contested Hong Kong market. The new calculus appears to be threefold:

1. Prioritize proprietary control: Banks are investing heavily in secure, in-house platforms like Arc to scale AI agents uniformly across global operations. This sidesteps the dependency on third-party tools that may be subject to sudden restrictions.
2. Manage tool-level risk: Access to third-party AI models is being reviewed on a tool-by-tool, jurisdiction-by-jurisdiction basis, reflecting a hyper-vigilant approach to data security and compliance.
3. Stay the course on automation: Despite the geopolitical headwinds, the drive to automate high-value tasks in a key financial center is not slowing down. If anything, the uncertainty is accelerating the shift toward controllable, internal AI factories.

This dynamic reflects a broader hardening of the US-China tech ecosystem, where partitions are deepening and corporate compliance is becoming as much about geopolitical strategy as it is about traditional regulation . For the banks navigating this new terrain, the ability to innovate securely isn't just a competitive advantage—it's a prerequisite for doing business at the fault line of a tech cold war.