JPMorgan Blocks Anthropic AI Access for Hong Kong Staff

JPM JPMorgan Chase has removed Anthropic's Claude AI models from its approved tools list for employees in Hong Kong, according to reporting by the Financial Times. The bank's decision follows a strict interpretation of Anthropic's terms of service, which sources say excludes use of Claude in Greater China, a designation that encompasses Hong Kong under the bank's compliance review. Hong Kong-based staff can no longer select Claude from JPMorgan's internal AI platform dropdown; other models remain available. Neither JPMorgan nor Anthropic has publicly confirmed the rationale, and the specific basis for the restriction is reportedly unconfirmed.

The move fits a broader pattern rapidly forming among global banks operating in Hong Kong. Goldman Sachs pulled Claude from its Hong Kong bankers' approved tools in late April 2026, citing a similar reading of its contractual arrangement with Anthropic. At the time, Goldman noted that models including Gemini and ChatGPT remained accessible on its internal platform, making the restriction Claude-specific rather than a blanket AI prohibition. Anthropic has separately tightened its usage policies to block Chinese-controlled entities from accessing Claude globally, and the US government has reportedly pressed the company to restrict exports of certain AI capabilities, adding to the compliance calculus for banks with operations in the region.

For JPMorgan, the practical effect is a reduction in AI tooling available to its Hong Kong workforce at a moment when banks are competing aggressively to embed generative AI into research, trading support, and back-office workflows. Anthropic is a private company with no public stock, but the restriction carries implications for JPMorgan's own AI strategy and competitive positioning in Asia. The episode also highlights the structural tension facing US AI providers: their terms of service increasingly reflect geopolitical constraints, which forces enterprise clients with cross-border operations to manage a patchwork of regional AI access policies rather than deploying tools uniformly.