Morning Edition · Thursday, November 12, 2026Published at 1:18 AM EST · New York
The revised Supported Regions Policy takes effect November 12. It covers Claude.ai, Claude Code, Cowork, the API and Claude models resold through cloud platforms, so checks on ownership and location now determine who can buy access to the model.

Anthropic's revised Supported Regions Policy takes effect today. The policy works as an allowlist, so any country or region not named on it is unsupported. Mainland China, Hong Kong, Macau and Russia are not on the list. The rules apply to Claude.ai, Claude Code, Cowork and the API, and to Claude models sold through cloud platforms and authorized resellers, according to the policy page and a summary from the Russian-language channel AI ML Big Data. The channel's main point is that employees of Russian and Chinese companies who work outside those countries can now fall under the restriction.
The policy sets three separate tests. The first bars anyone from using Claude while physically located in an unsupported region. The second bars any entity registered or headquartered in an unsupported region, along with its users and employees, regardless of where they are sitting. The third bars entities that are directly or indirectly majority-owned or controlled by people or entities in unsupported regions, wherever those entities operate, according to summaries by KuCoin and TokenPost.
Most of this builds on earlier rules. In September 2025, Anthropic barred companies more than 50% owned by entities in China, Russia, Iran or North Korea. At the time it said some groups were reaching Claude through subsidiaries incorporated in other countries. In July 2026, Bankless Times reported that Anthropic was moving against workarounds involving foreign subsidiaries, cloud platforms and VPNs. A NodeLoc forum analysis argues the new text mainly makes the location, registration and control rules explicit. The two clear additions are the control test, which goes beyond majority shareholding, and the employee test.
Chinese-language coverage has been heavier than Western coverage. Sina Finance called it the strictest wave of account bans yet. Secrss noted that Taiwan remains a supported region. Taiwan's ABMedia warned that a business trip to the mainland or Hong Kong could put an account at risk. One Chinese-language explainer reports that the policy also prohibits relaying requests through consumer subscriptions and keeps safety-monitoring data for 30 days. The desk could not match those two details against the official text.
Companies cut off from Claude still have alternatives. DeepSeek's V4 family is released under the MIT license, and Moonshot and Alibaba ship downloadable Kimi and Qwen weights, according to comparison trackers. The agentic-coding scores those trackers cite are mostly self-reported by the vendors, and the trackers do not always agree with each other.
Part of a tracked trend
Chinese Open-Weight Models Emerge as the Non-US AI Stack
As Washington restricts foreign access to US frontier models, governments and enterprises cut off from American AI increasingly standardize on downloadable Chinese open-weight models, splitting the world into competing AI supply blocs rather than a single frontier.
Start a discussion in Townsquare.
More from this edition
If this framing is true, Anthropic gains standing with Washington and with national-security buyers before its listing, and Chinese open-weight vendors such as DeepSeek, Alibaba and Moonshot gain the enterprise customers who are cut off.
The location, headquarters and control tests match ChainCatcher's reading and KuCoin's summary, but the desk found only secondary coverage of the November text. Anthropic's 2025 rationale is that Chinese law can compel companies to share data with intelligence services, and coverage does not resolve how cloud resellers will decide what counts as "control."
An open-source-intelligence read of how likely this story is true with its real nuance, not a judgment of any outlet. It assesses the claim, weighing independent and adversarial reporting. How we label confidence.
What this means
The policy changes how access is controlled. Checking an IP address is replaced by checking who owns the customer and where its users are physically located, and cloud resellers now have to enforce those checks themselves. Screening customers' ownership becomes a condition of buying model access, similar to the identity checks banks already run on new customers (desk thesis 6 on surveillance, and thesis 5 on computing capacity as a tool of geopolitical leverage). The most exposed are overseas subsidiaries of Chinese-controlled groups in Singapore, the Gulf and Europe that run Claude Code, the gray-market relay resellers operating inside China, and Amazon Web Services and Google Cloud, which now carry the cost of the ownership screening. Revenue loss for Anthropic is mostly already absorbed, because the 2025 rule had already removed direct Chinese corporate demand. The consensus treats this as a China ban, but the employee and control tests also reach mixed-ownership multinationals that do not identify as Chinese companies. The tracked export-control trend has been weakening, and here a private company is doing what Washington has not yet made law. The precedent is the October 2022 US chip controls, which sped up Huawei's substitution with its own Ascend chips and the Chinese open-weight push that followed. The likelier outcome is that Chinese-controlled multinationals shift coding-agent workloads to self-hosted Chinese open-weight models within months, which would expand the user base of those open-weight models outside China. This depends on those models staying close to Claude's performance on agentic coding. The read would be wrong if cloud resellers enforce the ownership tests loosely and relay access keeps running at scale.
What to watch
Observations to monitor, not financial advice.
Synthesized from: Polylog editors · Anthropic Supported Regions Policy · The Decoder · Sina Finance · Secrss
Comments
0No comments yet.