Anthropic is changing who can use Claude, and the new line is drawn around ownership as much as geography. The company will no longer provide services to companies that are majority-controlled by entities from China, Russia, Iran, or North Korea, citing national security concerns.
The change takes effect immediately and expands existing terms that already limited access in certain regions for legal and security reasons. The key shift is that a company outside a restricted region may still be blocked if its ownership structure links it back to one.
What Anthropic changed
The updated policy applies to entities directly or indirectly more than 50 percent owned by firms headquartered in unsupported regions. In practical terms, Anthropic is no longer looking only at where a customer is incorporated or where it signs in from. It is also looking at who controls that customer.
The source article says this is the first time a major US AI company has made such a policy change. The rule applies to China and to other countries the US classifies as adversaries, including Russia, Iran, and North Korea.
That matters because Claude is not being treated as a neutral software subscription in this policy. Anthropic is framing advanced AI model access as a security-sensitive capability, especially when the user is connected to a government or legal system that could compel cooperation with intelligence agencies.
The loophole Anthropic wants to close
Anthropic says companies from restricted regions like China have continued to access its services by setting up subsidiaries in other countries. The Financial Times points to a growing number of Chinese subsidiaries in Singapore that are used to circumvent controls and acquire US technology.
The concern is not only that a blocked company might use a different address. It is that a subsidiary can appear to be operating from an allowed market while remaining under the control of a parent company in a restricted region.
Under the new approach, that kind of ownership link becomes central. If the subsidiary is directly or indirectly more than 50 percent owned by a firm headquartered in an unsupported region, Anthropic’s policy can treat it as covered by the ban.
This makes the rule broader than a simple country block. It targets the corporate route around those blocks: a company structure that gives a restricted-region owner access through another jurisdiction.
Why Claude access is being treated as a security issue
Anthropic argues that firms controlled by authoritarian regimes can be compelled by law to share data and work with intelligence agencies. From the company’s perspective, that creates national security risks when those firms gain access to advanced AI models.
The source identifies two major concerns. First, companies could use AI capabilities to develop tools for rival military and intelligence agencies. Second, they could accelerate their own AI development through techniques like model distillation.
Model access is therefore not only about using Claude for ordinary business tasks. Anthropic is treating the models as capabilities that could be reused, studied, or embedded in broader technical programs by organizations subject to external state pressure.
The policy also reaches organizations that access Anthropic models via cloud services. That means the restriction is not limited to direct customers with an Anthropic contract. If a covered organization reaches Claude through another provider’s cloud offering, the same policy concern still applies.
Which companies could be affected
Major Chinese tech firms like ByteDance, Tencent, and Alibaba could be affected, according to the Financial Times. The source does not say that each company has been cut off, only that they could fall within the scope of the new policy.
The rule is based on ownership and control, so the practical impact depends on how a company or subsidiary is structured. The important threshold in the source is direct or indirect ownership of more than 50 percent by firms headquartered in unsupported regions.
For companies that rely on Claude directly, the consequence is straightforward: access may no longer be available. For companies using cloud services to reach Anthropic models, the policy signals that indirect access is also within scope.
An Anthropic executive acknowledged the company is likely to lose some business to competitors as a result. The financial impact on global revenue is estimated at a low hundreds of millions of dollars. Even with that cost, Anthropic says the move is necessary to address a significant problem.
What this means for AI competition in China
The source notes that US chatbots like Claude and ChatGPT are officially blocked in China, but can still be reached via VPNs. At the same time, China has developed a number of strong local alternatives, including Qwen, Deepseek, Kimi, and GLM.
That means the new Claude restriction may not immediately reshape everyday AI use in China. Chinese companies are likely to continue relying on homegrown solutions, at least while those tools remain practical for their needs.
The source also points to a condition that could make the restriction more meaningful over time. Anthropic’s new rules may only have a real impact in China once training runs on advanced AI accelerators, such as Nvidia’s banned chips, are significantly larger than those possible on domestic hardware.
Until then, the policy’s most immediate effect is likely to be on access pathways, compliance checks, and corporate structures. Anthropic is saying that where a company sits is no longer enough to determine whether it can use Claude. Who controls it now matters too.