Anthropic's Strictest Claude Access Rules Take Effect — China, Russia, and Chinese-Owned Entities Now Banned

Anthropic's most restrictive access policy to date took effect today, formally cutting off Claude across all surfaces — Claude.ai, the API, and cloud-distributed deployments — for users in mainland China, Hong Kong, Macau, and Russia. The policy also extends to any organization anywhere in the world that is majority-owned or controlled by individuals in those regions, regardless of where employees sit. It is the most sweeping access restriction Anthropic has implemented and closes loopholes that had allowed restricted-region users to reach Claude through third-party services.

The move reflects a broader tightening across the U.S. AI industry as frontier model developers come under pressure — from regulators, insurers, and their own security teams — to prevent advanced capabilities from reaching adversarial state actors or state-linked entities. Anthropic has banned 11.4 million accounts in 2026 for policy violations, though the new regional ban is structural, not reactive.

What the Policy Covers

The updated regional usage policy bans Claude access for:

• Individual users in mainland China, Hong Kong, Macau, and Russia
• Entities majority-owned or controlled — directly or indirectly — by individuals in those regions, even if the entity operates outside them
• Access surfaces covered: Claude.ai consumer products, Claude API, and models distributed through cloud partners (AWS Bedrock, Google Cloud Vertex AI)

The "indirect control" language is notable. It targets offshore holding structures and corporate shells that allowed restricted-region users to access Claude via nominally non-Chinese or non-Russian companies — a gap that prior policy versions left open. According to the policy, Anthropic identified three specific loopholes exploited before this version took effect.

Enforcement Mechanism

Anthropic's enforcement approach combines automated detection with identity verification requests for flagged accounts. Accounts that fail verification or are confirmed to fall under the new policy receive account closure notices. Cloud API access for affected accounts is revoked at the infrastructure level through coordination with AWS and Google.

Anthropic has not disclosed the detection methodology, which is standard — publishing it would assist circumvention. The 11.4 million account bans issued so far in 2026 suggest the enforcement pipeline has been running at significant scale ahead of today's policy formalization.

Why It Matters

For businesses with global operations, the indirect-control clause introduces compliance complexity. A multinational with a majority Chinese investor — even a passive one — may fall under the ban regardless of where it operates or who uses Claude day-to-day. Legal teams at companies with complex cap tables should audit their ownership structures against the new policy language.

For the AI industry, this sets a precedent. If Anthropic's access controls prove enforceable at scale, expect OpenAI, Google, and other frontier labs to face similar pressure to formalize comparable restrictions. The policy essentially treats Claude as a controlled export, mirroring the Commerce Department's chip export control logic.

For Anthropic specifically, the move carries commercial cost — China and Russia represent meaningful user populations — but positions the company favorably ahead of its IPO, which is currently in roadshow. Demonstrating robust geopolitical access controls is a material differentiator for enterprise and government buyers evaluating Claude for sensitive workloads.

What to Watch

Watch for how cloud partners AWS and Google handle affected enterprise accounts that purchased Claude access through their marketplaces before the policy took effect — contract and refund terms haven't been addressed publicly. Also watch for competing models based outside the U.S. (Mistral, Qwen, Deepseek) to market directly against this restriction as a feature, not a bug.