What exactly is happening

The Financial Times, citing sources familiar with the situation, reported that Anthropic is taking steps to close loopholes that allowed Chinese companies to access Claude models despite official restrictions. Among the firms that used these routes, according to the newspaper, is Ant Group, Alibaba's financial division operating in digital payments and fintech.

Anthropic has not officially commented on the details of its measures, but according to FT sources, these involve technical measures at the infrastructure and API level, designed to make it difficult or impossible to access from Chinese territory via intermediaries, proxy servers, and third-party cloud services.

Why Anthropic is blocking China

Anthropic, currently the world's most valuable AI startup with a valuation of around 965 billion dollars, has long built its identity on the responsible development of artificial intelligence. As early as 2024, it began restricting access to Claude from countries it considers risky — China was among the first on the list.

The reasons are multiple. Firstly, US export controls prohibit the provision of advanced AI models to entities in China without special permission. Secondly, there is a real concern that Chinese companies are using models like Claude for distillation — a technique where a smaller model "sucks up" the knowledge of a larger model by asking millions of questions.

After all, it wasn't long ago that Anthropic accused Alibaba of the largest distillation attack in history to date — according to researchers, the Chinese giant sent over 28.8 million queries to train its own model on Claude's responses. This incident clearly showed how intense the pressure is from Chinese labs to catch up with the American lead.

How Chinese firms circumvented restrictions

Despite the official blockade, Chinese firms found ways. Among the most common tricks were:

• API resellers — firms that purchased official access to Claude in permitted regions and then resold it to Chinese clients
• Cloud instances in Asia — using servers in Singapore, Japan, or South Korea as intermediaries
• VPN and proxy services — masking the user's true geographical location
• Corporate VPN networks — multinational companies used their global networks to allow Chinese branches to gain access

The last point likely applies to Ant Group — as a global fintech with branches outside China, it could have used its international infrastructure to access Claude.

Broader context: AI as a geopolitical weapon

Anthropic is not alone with its restrictions. OpenAI previously shut down Chinese accounts that misused ChatGPT for propaganda purposes. Similarly, Google restricts the availability of Gemini in certain regions. Moreover, the US government is escalating pressure — export restrictions on advanced AI chips from Nvidia for China are becoming increasingly strict.

The United States recognizes that a lead in artificial intelligence is a strategic advantage that cannot be relinquished. It's not just about economic competitiveness, but also national security — models capable of finding cyber vulnerabilities, generating convincing content, or analyzing vast amounts of data could be misused for attacks on critical infrastructure.

This is particularly relevant in the context of the Claude Mythos model, which Anthropic has deemed so capable in cybersecurity that it refuses to make it publicly available. If similar capabilities were to fall into the hands of actors linked to foreign states, the consequences could be serious.

What this means for Czech and European firms

For European businesses using Claude — and their number is growing in the Czech Republic — this news primarily means one thing: Anthropic closely monitors its platform. Firms that attempt to share access with colleagues in restricted regions risk losing their account.

At the same time, it strengthens the argument for European AI sovereignty, which is being discussed more and more loudly in Brussels. If American companies can restrict access to their models at any time for geopolitical reasons, Europe needs its own alternatives — whether it's projects like Mistral AI in France, Aleph Alpha in Germany, or emerging Czech initiatives in the field of language models.

Claude is officially available in the Czech Republic — while it doesn't support Czech as well as English, it is usable for generating Czech text, especially with newer versions like Claude Opus 4.8 or Claude Fable 5. Czech companies have particularly favored it for programming and technical documentation.

What's next

It can be expected that Anthropic will further tighten its security mechanisms. The commitment to "responsible AI" is a key part of its DNA — from its founding in 2021 by seven former OpenAI employees who disagreed with the company's commercial direction, to its current position as a leader in AI safety.

For Chinese firms, this means further complications in accessing the most advanced models. They already face restrictions on purchasing powerful chips, and if cloud AI services are systematically blocked as well, their ability to keep pace with American labs will further diminish. However, this also motivates Chinese researchers to accelerate the development of their own models — DeepSeek and other Chinese labs have already shown that they can compete even with limited resources.