Anthropic clarifies 3 grounds for restricting Claude access: an employee’s move abroad is not enough

A developer at a Russian company moves to Germany and expects to continue working in Claude Code. But being in a supported country does not, by itself, make that use permissible: Anthropic also considers the organization using the service. For a founder, this raises a practical question: can the product be built on the chosen provider at all, and how long would a migration take if access proved ineligible?

A revised version of Anthropic’s regional policy takes effect on November 12, 2026. It covers Claude applications, including Claude Code and Cowork, the API, and models accessed through cloud platforms and authorized intermediaries. But describing this as a sudden new ban on companies based abroad would be inaccurate: ownership restrictions were introduced earlier.

Three grounds to check

Anthropic’s regional policy lists three grounds for restricting access: a user’s physical presence in an unsupported region; an organization’s incorporation or headquarters in such a region; and direct or indirect majority ownership or control by individuals and organizations from unsupported regions. Restrictions on organizations also apply to their employees, even when those employees are in a supported country.

Consider a team that opens an office in Berlin but operates within a company incorporated in Russia. The office address alone does not settle eligibility. Similarly, a separate legal entity abroad requires a review of ownership and control. The reverse situation is also covered: an employee of an eligible organization cannot use the service while physically present in an unsupported region. A business trip does not create an exception.

There is also an important limit to how this should be interpreted. The wording on control by individuals from unsupported regions does not amount to an automatic ban based solely on a founder’s passport. The public policy does not provide sufficient grounds for that generalization. Ambiguous business structures should be discussed with the provider, with the actual circumstances clearly explained.

What changed, and why the date can be confusing

Anthropic announced on September 4, 2025 that the ban extends to companies controlled from unsupported jurisdictions, regardless of where they operate. In particular, it cited direct or indirect ownership of more than 50% by organizations headquartered in those regions. It cited national security risks and the potential use of its models to serve military and intelligence organizations as its rationale.

In its October 8, 2026 update announcement, Anthropic explicitly described the regional provisions as a clarification of how the existing policy applies. November 12 therefore marks the effective date of the current version. It does not establish that every foreign subsidiary had previously been free to use Claude without restriction.

That distinction matters for planning. The news does not demonstrate that mass account suspensions are beginning today. It shows which circumstances the provider considers grounds for denying access. The published document contains neither a detailed procedure for checking every customer nor a blocking schedule. Telling readers they will immediately lose their accounts based on this page alone would be speculation.

What teams with Claude embedded in their workflows should do

Our practical recommendation: start by mapping dependencies. Where is Claude used—in a developer’s editor, document processing, customer support, or an agent with access to tools? Who signed the contract, where does the team work, and who controls the company? These questions need answers before the next expansion of the integration, because a working API key says nothing, by itself, about whether that use is permitted.

Next, test whether another model can replace it on real tasks. If Claude extracts data from contracts, run an alternative on the same documents and compare errors, cost, and latency. If it operates tools, test the entire workflow: a different model may need different instructions, and the results of its actions may need different checks. Simply changing the API endpoint is sometimes insufficient.

The product manager should decide in advance what customers will see if the model becomes unavailable: a task queue, a temporarily simplified feature, or a clear message explaining the delay. The founder should clarify the organization’s eligibility. The lesson for managing dependence on a provider is that regional terms belong in a platform assessment alongside response quality, price, and speed.