The episode highlights a practical issue for multinational companies: an enterprise account does not necessarily make an AI service available from every location where employees work or travel.

What triggered OKX’s restriction?

Bloomberg reported that OKX stopped providing Claude access to employees in the region after Anthropic temporarily suspended the company’s corporate account. The account was subsequently restored, but OKX retained the location-based restriction.

The reported sequence was:

1. Anthropic briefly suspended OKX’s corporate Claude account in early August.
2. OKX identified employee use in Hong Kong that may not have complied with Anthropic’s geographic access policies.
3. The exchange blocked Claude for Hong Kong-based employees and for staff traveling through mainland China.
4. Affected employees were directed to other AI models instead.

OKX said it spends between $6 million and $8 million per month across large-language-model providers. That figure covers multiple providers; the available reporting does not establish how much OKX spent specifically on Claude.

What did the internal notices reportedly say?

The notices reportedly told employees that Anthropic did not permit Claude use in Hong Kong or mainland China. They also instructed staff not to bypass the geographic restriction with a VPN and to route Claude-related requests to alternative AI models.

The precise wording of the internal messages cannot be independently verified from the supplied excerpts, so those details should be treated as reported guidance rather than a fully reproduced company policy. The consistent point across the reporting is that OKX treated location as a condition of Claude access, including when employees were traveling.

That approach also explains why restoring the corporate account did not automatically restore access for every employee. Account-level availability and location-level eligibility were separate issues.

How does the decision compare with Goldman Sachs?

Goldman Sachs took a similar step in April, removing Claude access for bankers based in Hong Kong and for bankers visiting the territory. Reuters reported that employees had previously accessed Claude through the bank’s internal AI platform but no longer could.

The two cases are not identical. OKX’s restriction followed a temporary corporate-account suspension and reported concerns about use in unsupported locations. Goldman’s case was described in reporting as a territorial-access or contractual issue, while the bank maintained its broader relationship with Anthropic.

Together, the cases show that a company may continue working with Anthropic globally while limiting Claude access for employees in Hong Kong. Geographic eligibility can override an otherwise broad enterprise relationship.

What these cases reveal about Claude access in Hong Kong and mainland China

The available reporting supports a narrow but important conclusion: Claude use from Hong Kong and mainland China was treated as outside Anthropic’s permitted geographic coverage in these corporate cases.

That creates operational challenges for companies with mobile or distributed workforces. Employees may move between supported and unsupported locations, while centrally managed AI tools can be accessed through corporate platforms, APIs, or internal applications. A provider’s regional rules therefore become part of ordinary software-compliance planning.

The reporting does not establish a definitive policy for every China-linked relationship. It remains unclear from the supplied evidence whether the restrictions apply in the same way to mainland China-based clients, customers, business partners, or other counterparties. The documented cases concern employee use from specific locations, not every possible form of commercial interaction.

The broader business significance

The restrictions show how geopolitical and regional-access questions can reach day-to-day corporate workflows. In Hong Kong’s financial and technology sectors, companies may want to expand AI use while still managing provider eligibility, account continuity, data-security concerns, and employee travel.

For enterprise buyers, the practical lesson is straightforward: evaluate AI tools by more than model capability or price. A deployment review should also ask where the service is officially available, whether access changes by user location, how travel is handled, what bypasses are prohibited, and which fallback models are available if an account is interrupted.

OKX’s experience illustrates the risk clearly. Anthropic restored the company’s account, but the exchange still chose a more restrictive internal policy to avoid another disruption and route affected work to alternative models.

The episode highlights a practical issue for multinational companies: an enterprise account does not necessarily make an AI service available from every location where employees work or travel.

What triggered OKX’s restriction?

Bloomberg reported that OKX stopped providing Claude access to employees in the region after Anthropic temporarily suspended the company’s corporate account. The account was subsequently restored, but OKX retained the location-based restriction.

The reported sequence was:

1. Anthropic briefly suspended OKX’s corporate Claude account in early August.
2. OKX identified employee use in Hong Kong that may not have complied with Anthropic’s geographic access policies.
3. The exchange blocked Claude for Hong Kong-based employees and for staff traveling through mainland China.
4. Affected employees were directed to other AI models instead.

OKX said it spends between $6 million and $8 million per month across large-language-model providers. That figure covers multiple providers; the available reporting does not establish how much OKX spent specifically on Claude.

What did the internal notices reportedly say?

The notices reportedly told employees that Anthropic did not permit Claude use in Hong Kong or mainland China. They also instructed staff not to bypass the geographic restriction with a VPN and to route Claude-related requests to alternative AI models.

The precise wording of the internal messages cannot be independently verified from the supplied excerpts, so those details should be treated as reported guidance rather than a fully reproduced company policy. The consistent point across the reporting is that OKX treated location as a condition of Claude access, including when employees were traveling.

That approach also explains why restoring the corporate account did not automatically restore access for every employee. Account-level availability and location-level eligibility were separate issues.

How does the decision compare with Goldman Sachs?

Goldman Sachs took a similar step in April, removing Claude access for bankers based in Hong Kong and for bankers visiting the territory. Reuters reported that employees had previously accessed Claude through the bank’s internal AI platform but no longer could.

The two cases are not identical. OKX’s restriction followed a temporary corporate-account suspension and reported concerns about use in unsupported locations. Goldman’s case was described in reporting as a territorial-access or contractual issue, while the bank maintained its broader relationship with Anthropic.

Together, the cases show that a company may continue working with Anthropic globally while limiting Claude access for employees in Hong Kong. Geographic eligibility can override an otherwise broad enterprise relationship.

What these cases reveal about Claude access in Hong Kong and mainland China

The available reporting supports a narrow but important conclusion: Claude use from Hong Kong and mainland China was treated as outside Anthropic’s permitted geographic coverage in these corporate cases.

That creates operational challenges for companies with mobile or distributed workforces. Employees may move between supported and unsupported locations, while centrally managed AI tools can be accessed through corporate platforms, APIs, or internal applications. A provider’s regional rules therefore become part of ordinary software-compliance planning.

The reporting does not establish a definitive policy for every China-linked relationship. It remains unclear from the supplied evidence whether the restrictions apply in the same way to mainland China-based clients, customers, business partners, or other counterparties. The documented cases concern employee use from specific locations, not every possible form of commercial interaction.

The broader business significance

The restrictions show how geopolitical and regional-access questions can reach day-to-day corporate workflows. In Hong Kong’s financial and technology sectors, companies may want to expand AI use while still managing provider eligibility, account continuity, data-security concerns, and employee travel.

For enterprise buyers, the practical lesson is straightforward: evaluate AI tools by more than model capability or price. A deployment review should also ask where the service is officially available, whether access changes by user location, how travel is handled, what bypasses are prohibited, and which fallback models are available if an account is interrupted.

OKX’s experience illustrates the risk clearly. Anthropic restored the company’s account, but the exchange still chose a more restrictive internal policy to avoid another disruption and route affected work to alternative models.