Anthropic tightens access restrictions on its AI
Anthropic is expanding its restrictions to block access for companies controlled from non-permitted regions, even when they operate through subsidiaries in other countries. The new policy covers entities with more than 50% ownership or indirect control from jurisdictions such as China.
Anthropic expanded its restrictions to prevent companies linked to non-permitted countries, such as China, from accessing its services even when they operate from another country.
The company announced the change on September 4, 2025. Until now, its terms of service had already prohibited the use of its products in certain regions for legal, regulatory and security reasons. The problem, according to Anthropic, is that some companies were bypassing the restriction through subsidiaries registered in permitted countries.
The new rule also looks at who controls the company
The update no longer focuses only on where a company is incorporated. It also considers its ownership structure and the control it may receive from a restricted jurisdiction.
The rule covers entities that are more than 50% controlled or owned, directly or indirectly, by companies headquartered in non-permitted regions. China is the main example, although the rule applies to any jurisdiction included in Anthropic's restrictions.
The reason is that a subsidiary does not always operate as an independent company. Anthropic argues that companies subject to the influence of authoritarian governments may be forced to share data, cooperate with intelligence services or comply with other government demands, regardless of where the subsidiary is registered or what its employees prefer.
Why Anthropic considers this important
The company's concern is that its models could be used to develop tools for the military or intelligence services of countries considered adversaries. It also points to another risk: these models could help accelerate the development of proprietary systems through techniques such as distillation, a process that trains another model to imitate part of the behavior of a more advanced one.
In practice, a company headquartered in a permitted country could lose access if its parent company or owners are subject to control from a restricted region. This could affect subsidiaries, technology providers and international groups with complex ownership structures.
The measure is not a new law or a general ban imposed on every company from a particular country. It is a change to Anthropic's commercial terms. The company may assess whether an organization meets its requirements before granting access or continuing to provide its services.
Part of a broader policy
Anthropic also reiterated its support for other measures related to national security and technological competition:
- Stricter export controls to limit authoritarian governments' access to advanced systems.
- More energy and infrastructure projects in the United States to support AI growth.
- Rigorous model evaluations to detect capabilities that adversaries could exploit.
For you, the direct impact will depend on the company you use. If you work with an international company, its business location will no longer be the only factor: its owners, parent company and control relationships will also matter. The next point to watch will be how Anthropic and other AI companies apply these rules to multinational groups operating across several countries.