OKX restricts Claude in Hong Kong after Anthropic briefly suspended its account

OKX has restricted employees based in Hong Kong, as well as staff travelling through mainland China, from using Anthropic’s Claude artificial-intelligence model after the crypto exchange’s corporate account was briefly suspended earlier in August 2026.

The move, reported by Bloomberg on August 20, comes months after Goldman Sachs imposed a similar restriction on its Hong Kong staff, making the OKX Claude restrictions part of a wider shift in how global companies manage access to US-developed AI models.

OKX stopped providing Claude access to affected employees after the suspension, according to people familiar with the matter and internal messages reviewed by Bloomberg. Anthropic later restored the exchange’s corporate account, but OKX has chosen to maintain the regional restriction and direct affected employees toward alternative AI models.

The OKX Claude restrictions do not amount to a company-wide termination of Claude. Employees operating in jurisdictions where Anthropic supports the service can continue to use the model through OKX’s restored enterprise account.

OKX CEO Star Xu addressed the situation publicly on August 19, saying: “We appreciate Anthropic’s constructive engagement as we work together to address this matter and restore our account access.” — Star Xu, CEO, OKX.

The exchange has also indicated that it will not facilitate the use of virtual private networks to circumvent Anthropic’s geographic restrictions while employees are travelling through affected jurisdictions.

Goldman Sachs faced similar Claude restrictions earlier

The OKX Claude restrictions follow a precedent set by Goldman Sachs, which removed Claude access from its Hong Kong-based bankers in April.

Reuters reported on April 29 that Goldman employees in Hong Kong had previously been able to access Claude through the bank’s internal AI platform but lost access in the weeks before the report. Other AI systems, including ChatGPT and Google Gemini, remained available to Goldman employees.

The Goldman decision followed a strict interpretation of the bank’s contract with Anthropic after consultation with the AI company, according to Reuters. Goldman concluded that employees in Hong Kong should not have access to Anthropic’s products under the agreement.

An Anthropic spokesperson told the Financial Times that its Claude models had never been officially “supported” in Hong Kong, although the company declined to comment further.

That clarification is important to the OKX Claude restrictions because Hong Kong occupies a distinctive position in the international technology market. While it operates separately from mainland China in many areas, Anthropic does not list Hong Kong among the markets where Claude and its API are officially accessible.

The Goldman Sachs case also showed that companies with global AI agreements cannot necessarily assume that an enterprise contract permits employees to access a model from every location where the company operates.

JPMorgan adds to growing restrictions on Anthropic models

The OKX Claude restrictions are also part of a broader pattern among major financial institutions.

In June, JPMorgan Chase restricted Hong Kong employees from accessing Anthropic’s AI models, including Claude, from its internal approved-tools list. Reuters reported that the decision was linked to concerns over the wording of Anthropic’s licensing agreement and followed Goldman Sachs’ earlier action.

The sequence of decisions means that restrictions have now affected employees at major institutions spanning cryptocurrency and traditional finance.

For OKX, however, the immediate trigger was different. While Goldman Sachs and JPMorgan’s restrictions were linked to contractual and licensing interpretations, the OKX Claude restrictions followed the temporary suspension of the exchange’s corporate Anthropic account.

Bloomberg reported that OKX acknowledged that some previous access by employees may not have complied with Anthropic’s regional policies. The exchange has since restored its account but tightened internal controls around where Claude can be accessed.

The distinction suggests that companies are increasingly required to monitor not only who is using an AI service but also where that person is physically located when accessing it.

AI spending raises stakes for OKX

The OKX Claude restrictions are particularly significant because the exchange has made artificial intelligence a major part of its internal operations.

Star Xu said OKX spends between $6 million and $8 million every month across multiple leading AI model providers. That would represent an annualised expenditure of roughly $72 million to $96 million if spending remains at the same level.

The exchange has also developed its own internal AI platform, Oli, which combines multiple large language models and development tools. OKX has said AI is deeply integrated into its engineering workflow, meaning restrictions affecting one model do not necessarily stop its broader AI operations.

The company can instead route affected employees to alternative models, reducing its dependence on Anthropic.

Still, the OKX Claude restrictions demonstrate the growing complexity surrounding enterprise AI adoption. A company can maintain a global workforce and a corporate AI contract while still needing to prevent employees from accessing a particular model in specific jurisdictions.