Crypto exchange OKX has restricted employees in Hong Kong — and staff traveling through China — from using Anthropic's Claude AI assistant, after the company's corporate Anthropic account was briefly suspended in early August, Bloomberg reported. The episode, confirmed in details published across multiple outlets on August 19 and 20, 2026, highlights an uncomfortable new reality for global enterprises: the AI models that have become critical workplace infrastructure come with regional licensing boundaries that do not match how multinational companies actually operate.
Access to OKX's account has since been restored. But according to the reports, the exchange will now prevent affected employees from routing requests through Claude and will direct their work to alternative AI models instead. OKX said it recently became aware that some employees' Claude usage in Hong Kong may not have fully complied with Anthropic's regional access policies. For more context on this story, see our ongoing AI trends.
The Policy Behind the Suspension
The root cause is straightforward: Anthropic does not permit Claude access in either Hong Kong or mainland China, according to the company. That policy puts the territory — home to one of Asia's densest concentrations of banks, crypto firms and trading houses — outside Claude's official footprint, even as rivals like OpenAI and Google race to sign enterprise deals across the Asia-Pacific region.
OKX is not alone in hitting this wall. Goldman Sachs previously removed Claude from the AI tools available to its Hong Kong employees after determining that its Anthropic agreement did not cover use in the territory. JPMorgan subsequently restricted Claude access for its own Hong Kong staff after reviewing similar licensing terms. In each case, the banks' global contracts collided with a provider that treats Hong Kong as a restricted jurisdiction.
$6 Million a Month on AI Models
The scale of OKX's AI dependence makes the restriction more than a footnote. The exchange disclosed that it spends between $6 million and $8 million every month across multiple AI models — an annualized run rate of roughly $72 million to $96 million if spending holds at current levels.
OKX operates globally, with major offices spanning Hong Kong, Singapore, Dubai, London, California, New York, Malta and Istanbul. The company has been recruiting senior AI personnel, including advertising for a senior director or vice president of AI covering Singapore and Hong Kong during 2026.
That spending pattern, and the multi-vendor strategy behind it, is what allowed OKX to absorb the disruption. Because the exchange uses several large language models rather than depending exclusively on Claude, it could redirect Hong Kong employees toward alternative systems that are permitted in the jurisdiction.
Geographic Resilience Becomes an AI Strategy
For multinationals, the lesson is sinking in: maintaining a diversified set of AI providers now carries an additional strategic benefit — geographic resilience. Regional restrictions imposed by one vendor do not necessarily block employees from accessing competing models permitted in the same territory.
The dynamic illustrates an increasingly complicated operating environment for international financial and technology companies. US-developed AI models are becoming core enterprise infrastructure at precisely the moment Washington and Beijing are tightening controls around technology transfer and data flows. Providers, for their part, are drawing their own compliance maps — and Hong Kong, with its unique regulatory status, has landed on the wrong side of Anthropic's.
There is also a commercial dimension to the story. Anthropic has been aggressively courting enterprise customers with privacy and compliance commitments as it competes with OpenAI for corporate accounts. But a provider's compliance map can undo its sales pitch: for any firm whose workforce is anchored in Hong Kong, Claude is effectively off the menu regardless of how well it performs.
What Comes Next
OKX said it appreciated Anthropic's engagement in resolving the corporate-account suspension and indicated that its response was designed to ensure compliance with model providers' policies. Anthropic has not publicly said whether its regional access policies might change, and the company's stated policy still excludes both Hong Kong and mainland China.
The competitive stakes are high. Anthropic has been courting enterprise customers with privacy and compliance commitments — commitments that prompted OpenAI to respond this week with its own zero data retention options for frontier models, as the two rivals trade blows for corporate accounts. But the Hong Kong episode shows the limits of Anthropic's enterprise push: a bank or exchange with a major hub in the territory cannot standardize on Claude, no matter how strong the model's coding or analysis capabilities.
For now, the pattern appears to be repeating across the industry: firms discover a licensing gap, suspend or restrict access in the affected region, and rebalance their AI portfolios. As AI spending moves from experiment to line item — measured at OKX in the millions per month — expect procurement teams to start treating model access maps with the same seriousness as data residency and sanctions compliance. The era when a single AI subscription could serve every office a company holds is over.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
