Anthropic has begun blocking Chinese companies from using its Claude AI models, following an investigation by the Financial Times that exposed how major firms like Ant Financial and ByteDance were using overseas subsidiaries and cloud infrastructure to evade U.S. export controls. The move, implemented in recent days without a public announcement, marks a significant step in the tech industry’s compliance with Washington’s tightening restrictions on advanced AI technology reaching China.

According to the Financial Times, Ant Financial—an affiliate of Alibaba—and ByteDance, the parent of TikTok, were among those that accessed Claude through workarounds. Ant Financial reportedly used its international entity Ant Group on Amazon Web Services (AWS) in the United States, while ByteDance operated via a Singapore-based subsidiary on Google Cloud. Both companies allegedly leveraged Claude’s capabilities to develop internal tools and compete with Western AI services, putting Anthropic in a precarious legal and reputational position.

The enforcement action is not an isolated event. It reflects a broader crackdown by U.S. AI companies, long under pressure to ensure their technologies do not inadvertently fuel China’s military or surveillance ambitions. The clampdown also follows a series of warnings from the Biden administration that cloud service providers must step up identity verification to prevent restricted nations from accessing sensitive AI models. For the many Windows users and IT professionals who integrate Claude into their workflows or custom applications, the incident is a stark reminder of how geopolitics can instantly reshape the AI toolbox.

How the Block Was Detected

Anthropic’s trust and safety team appears to have rolled out automated controls that flag accounts based on IP geolocation, billing country, and usage patterns indicative of relay access. Accounts found to be originating from China—even when masked by overseas billing addresses—are being terminated or suspended, often without prior notice. The company’s acceptable use policy, last updated earlier this year, explicitly prohibits use in countries subject to U.S. embargoes, but enforcement had been lax until the FT report brought the issue to light.

The shutdowns have not been limited to large enterprises. Individual developers and small startups in China who depended on the Claude API for prototyping or production apps have also reported sudden loss of service on social media and developer forums. Some of these developers were using Claude to power Windows desktop applications or browser extensions, and must now scramble to integrate alternative large language models.

Who’s Affected—and How

Chinese Enterprises and Startups
The most immediate disruption hits Chinese tech firms that had embedded Claude into their products. For instance, ByteDance was reportedly using Claude to train its own AI models and to power content moderation tools. Ant Financial used it for chatbot development and data analysis. With access severed, these projects face delays unless the companies successfully pivot to domestic alternatives—a process that can take weeks or months given the need to retrain models and adjust integrations.

Multinational Corporations
Global companies with legitimate operations in China are also feeling the sting. Many use a single corporate AWS account to access AI services across multiple regions, and employees traveling in China may find their Claude access blocked even if the billing entity is based outside the country. IT departments must now navigate a complex compliance landscape, ensuring that Chinese subsidiaries do not accidentally violate U.S. export laws. For Windows-centric enterprises that rely on Azure or AWS for AI workloads, this means re-evaluating endpoint management and VPN configurations.

Independent Developers and Windows Users
While most everyday Windows users are unlikely to notice a direct impact, those who build and sell Windows applications that leverage Claude’s API—through Electron apps, chatbots, or productivity tools—will need to re-engineer their backends. Popular AI-powered Windows apps like coding assistants, writing tools, and design plugins that rely on Claude could lose functionality for Chinese users, potentially fragmenting the user base. Developers are advised to keep their AI integrations model-agnostic and to monitor Anthropic’s developer portal for any grace periods or appeals processes.

The Path to This Point: A Timeline of U.S.-China AI Tensions

To understand why Anthropic is acting now, it helps to trace the escalating U.S. restrictions on technology exports to China.

  • October 2022: The U.S. Commerce Department’s Bureau of Industry and Security (BIS) imposes sweeping controls on advanced semiconductor exports to China, targeting chips essential for AI training.
  • January 2023: Reports emerge of Chinese entities using cloud services to bypass chip restrictions, prompting lawmakers to call for tighter cloud provider regulations.
  • October 2023: The BIS expands chip controls to cover a wider range of AI accelerators and adds enforcement rules for “deemed exports”—when technology is released to a foreign national within the United States.
  • Early 2024: The Biden administration considers a rule requiring cloud providers to verify the identities of their foreign customers, but the proposal stalls in interagency review.
  • Spring 2024: OpenAI, Google, and Anthropic update their terms of service to more explicitly ban Chinese access, but enforcement remains patchy. Chinese firms continue to access AI models through intermediaries.
  • November 2024 (estimated): The Financial Times publishes its investigation detailing how Ant Financial and ByteDance circumvented Anthropic’s geoblocking. Within days, Anthropic initiates mass account terminations.

Throughout this period, Chinese domestic AI companies have been racing to close the gap with Western models. Baidu’s Ernie, Alibaba’s Tongyi Qianwen, and Moonshot AI’s Kimi have all demonstrated capabilities that rival GPT-4 and Claude 2 on several benchmarks. However, Claude has maintained an edge in safety, chain-of-thought reasoning, and long-context understanding—features that many Chinese firms considered worth the compliance risk.

What to Do Now: Practical Steps for Affected Users

If your access to Claude has been interrupted, here are the concrete steps you should take today.

1. Confirm the Block and Check for Recourse

First, log into your Anthropic developer account and check for any messages regarding suspension. If you believe your account was mistakenly flagged—for example, you are a non-Chinese entity with occasional Chinese logins—file an appeal with Anthropic’s support, providing documentation that proves your primary place of business. Be aware that appeals may take weeks to resolve.

2. Switch to Chinese AI Models

For most users in China, the fastest path to restoring functionality is to adopt a domestic model. These options are not affected by U.S. export controls and are optimized for Chinese-language applications:

  • Baidu Ernie 4.0 – A multimodal model with strong language understanding, available via Baidu’s cloud platform.
  • Alibaba Tongyi Qianwen – Popular among enterprises already on Alibaba Cloud; offers API access similar to Claude’s.
  • Zhipu AI GLM-4 – An open-source alternative with strong commercial support.
  • Moonshot AI Kimi – Known for handling extremely long contexts, much like Claude’s 200K token window.

Many of these models provide Python SDKs and REST APIs that are compatible with common AI frameworks like LangChain, making the switch less painful than a ground-up rebuild.

3. Re-engineer Your Application for Model Flexibility

Rather than locking your Windows app or backend service to a single provider, build an abstraction layer that can route requests to multiple models. Tools like LiteLLM or custom middleware can automatically fall back to alternative services when one is unavailable. This not only insulates you from future geopolitical shocks but also lets you take advantage of new models as they emerge.

4. Strengthen Organizational Compliance

For IT administrators managing global teams, now is the time to audit all AI service subscriptions. Identify any accounts that may be using Chinese billing addresses or that regularly connect from mainland IP addresses. Implement network-level logging on your Windows endpoints to detect unauthorized AI API calls. If your organization has a presence in China, consult legal counsel to establish data flows that comply with both U.S. and Chinese regulations.

Outlook: A Balkans-Inducing AI Landscape

Anthropic’s crackdown is unlikely to be the last. As geopolitical tensions persist, other U.S.-based AI companies—including OpenAI, Cohere, and even Microsoft itself—will feel compelled to follow suit, potentially closing similar loopholes. Microsoft’s Azure AI services already block users from certain regions, but the company’s partnership with local Chinese cloud providers like 21Vianet might create gray areas that will come under scrutiny.

The longer-term consequence is almost certain: a bifurcated AI ecosystem where Chinese and Western models develop along separate trajectories, with little interoperability. For Windows developers and IT professionals, this means supporting two parallel toolchains—one for the domestic Chinese market and one for the rest of the world. It also means that the days of one-size-fits-all AI strategies are over.

Watch closely for any announcements from the U.S. government regarding mandatory cloud customer verification rules. If finalized, such rules would dramatically raise the compliance burden for all cloud AI providers and could even affect individual developers who share accounts across borders. In the interim, the message from Anthropic is clear: if you’re in China, Claude is off the table.

The article was informed by reporting from the Financial Times and background on U.S. export controls from the Bureau of Industry and Security.