Anthropic to Cut Off Claude for Russian Companies From November 12, Even Those With Foreign Legal Entities

If your team writes code with Claude Code or runs production through the Anthropic API, and the company is backed by Russian owners, you now have an urgent engineering task. The updated Supported Regions policy takes effect on November 12, 2026, and "we have a legal entity in Kazakhstan or the UAE" stops being a workable answer.

What exactly is changing

According to CNews, access is being closed to organizations registered in Russia and to companies controlled by Russian persons. The country of registration does not matter: the criterion is control.

The restriction covers the entire product line:

• the Claude.ai chat;
• Claude Code, the development tool;
• the API;
• Claude models available through cloud platforms and resellers.

The last item is the most painful. Many teams did not have a direct contract with Anthropic and instead got the models through a cloud provider or an intermediary. Now that loophole is closed in the policy itself: resale does not cancel the rules.

Imagine a startup with an office in Dubai, Russian founders and a product built on Claude. Previously it was formally a customer from a "supported region." Now the deciding question is who owns and manages it.

Why Anthropic says this isn't news

The company stresses that this is not a new ban but a clarification of previous rules. The region was never on the list of supported ones; the wording simply allowed interpretations that customers exploited via third countries.

The motive, as the source describes it, is commercial. Anthropic is preparing for an IPO and needs to exclude from its revenue customers who violate the regional policy. Ahead of a public offering, investors and auditors look at the quality of revenue. Income from users who should not be there under the rules becomes a legal and reputational risk. It is easier to cut it off in advance.

For the industry, this is a signal: major model providers are becoming stricter on compliance as they mature and head to the stock market. Formal workarounds, such as setting up a company in a convenient jurisdiction, work less and less well with market leaders.

What teams should do

There is no need to panic, but an audit should not be postponed. Here is a minimal plan of action:

• Map your dependencies. Where Claude is used in the product and internal processes: direct API, cloud provider, IDE plugins, contractor integrations.
• Check your ownership and control structure. If you fall under the criterion, assume access will disappear rather than "somehow get extended."
• Build an abstraction layer over the model. If prompts and calls are tied to a single vendor, migrating becomes a refactoring project. A thin adapter layer and a set of regression tests for response quality greatly reduce the cost of switching providers.
• Test alternatives in advance: models from other providers and open models you can self-host. Quality on your tasks will have to be measured anew.
• Legal questions need a lawyer, not a Telegram chat. What counts as "control" and how it will be checked in practice is determined by the text of Anthropic's policy.

Limitations and what remains unclear

The source does not disclose how verification will work: how Anthropic will identify ultimate beneficiaries, how strictly it will track resellers, and what will happen to already paid subscriptions and accumulated data. These details will become clear once actual blocks begin.

The main takeaway: dependence on a single closed vendor can now be considered not only a technological risk but also a political and legal one. Teams in affected jurisdictions should allow time for migration and keep a fallback option ready before the account is blocked.