🌐 🇵🇱 Polski · 🇬🇧 EN
On July 1, 2026, China introduced restrictions on the export of open artificial intelligence models, triggering global repercussions. Why did Beijing make this decision, and how will it impact technology development, scientific research, and the balance of power in the AI race?
New regulations: what exactly changed in 2026?
As of July 1, 2026, China's Ministry of Commerce (MOFCOM) and the General Administration of Customs (GACC) have implemented updated export regulations that significantly restrict the ability to share open AI models outside the country's borders. These regulations are a response to rising geopolitical tensions and national security concerns, but their consequences extend far beyond Chinese companies.
Key assumptions of the regulations
- Definition of "open AI models": The regulations cover models released under open-source licenses (e.g., Apache 2.0, MIT) or in "open-weight" format that could be used for military purposes, surveillance, or destabilization. It is no longer just about closed systems – now, projects that were previously available to everyone are also under control.
- Restriction criteria:
- Models with a parameter count above 10 billion (in 2023 the threshold was 100 billion, and in 2025 it was lowered to 50 billion).
- Models trained on datasets containing sensitive information, such as data on critical infrastructure or personal data of Chinese citizens.
- Models utilizing dual-use technologies (civilian-military), e.g., for satellite image analysis or natural language processing in a security context.
- Export procedure:
- Requirement to obtain an export license for every transaction or release of a model outside of China.
- Obligation to report all model versions (including updates) to MOFCOM within 30 days of their publication.
- Ban on exports to countries under UN or Chinese sanctions (e.g., North Korea, Iran).
- Penalties for violations: Fines of up to 5 million yuan (approx. 700,000 USD) and a 5-year ban on conducting export activities.
Why now?
The Chinese government's official justifications focus on three main arguments:
- National security: In a MOFCOM statement from May 15, 2026, it was emphasized that open AI models could be used by "foreign forces" to interfere in China's internal affairs or for industrial espionage. A quote from the document: "AI technologies are a key element of digital sovereignty and cannot be subject to uncontrolled transfer."
- Intellectual property protection: Chinese companies invest billions of yuan in AI model development, and their uncontrolled export could lead to technology theft by competitors, especially from the United States.
- Geopolitical balance: China argues that the restrictions are a response to American export limitations (e.g., the ban on selling NVIDIA chips to China between 2022–2025) and the European 2024 AI Act, which imposes obligations on model providers.
Which projects and companies are most affected?
The restrictions have primarily hit Chinese open-source projects that were previously widely used around the world. Here are the most important ones:
Withdrawn or restricted models
- Qwen (Alibaba): The Qwen2-72B and Qwen2-VL models were removed from public GitHub repositories on July 2, 2026. Alibaba issued a statement committing to compliance with local laws but did not provide details regarding the project's future.
- Yi (01.AI): The Yi-1.5-34B model disappeared from the Hugging Face platform on July 3, 2026. The company announced it would only make it available to "trusted partners" after obtaining the appropriate licenses.
- deepseek (deepseek-AI): The deepseek-V2 (236 billion parameters) and DeepSeek-Coder models were subjected to restrictions. DeepSeek-AI announced it is suspending exports to countries not on the "whitelist."
- chatglm (Zhipu AI): The smaller ChatGLM3-6B model (6 billion parameters) is not covered by the restrictions, but its larger versions have been withdrawn.
Reactions of Chinese tech giants
- Alibaba Cloud: Announced the creation of a "Chinese open-source AI zone," where models will be available only to entities registered in China. This solution aims to formally comply with regulations while maintaining control over access to the technology.
- Baidu and Tencent: Restricted access to their models (Ernie Bot, Hunyuan) for users outside of China. This means that even existing integrations may stop working for foreign clients.
Global consequences: disruptions in the AI ecosystem
Chinese restrictions have far-reaching effects on the global open-source AI ecosystem. Here are the most significant ones:
Disruptions in international projects
- Hugging Face: The platform removed over 50 Chinese models, including the popular Qwen and Yi. In a statement on July 4, 2026, the company expressed concern over the "fragmentation of the AI ecosystem" but emphasized that it must comply with local laws.
- GitHub: Chinese repositories (e.g., Qwen, Yi) have been marked as "restricted" for users outside of China. Microsoft, the owner of GitHub, has not commented on the matter, but these changes are visible to anyone attempting to access these projects.
Impact on scientific research
- Universities: Institutions such as Tsinghua University and Peking University have suspended cooperation with foreign partners regarding model sharing. This may lead to the isolation of Chinese science in key research areas.
- Scientific publications: An article in the journal Nature from July 10, 2026, warns of a "new iron curtain in AI research," which could hinder knowledge exchange and slow down scientific progress. Has artificial intelligence become a "second intelligence"? – this question takes on new meaning in the context of these changes.
- Scientific conferences: Organizers of conferences such as NeurIPS 2026 are considering introducing "censorship-free zones" where Chinese researchers could present results without restrictions. However, this raises questions about the balance between academic freedom and compliance with local laws.
Alternative solutions
In response to Chinese restrictions, the open-source community is looking for ways to bypass the limitations:
- Decentralized repositories: Projects like Gitopia or Radicle are gaining popularity as alternatives to centralized platforms. They allow for storing and sharing code without the mediation of companies subject to Chinese jurisdiction.
- Mirroring tools: The Shotgun project (available on GitHub) enables the creation of copies of Chinese models outside of China. This solution is controversial, however, as it may violate technology export regulations.
- Development of local models: Companies from the US and EU, such as Mistral AI or Stability AI, are accelerating work on models independent of China. This could lead to the emergence of new industry leaders, but also to increased costs and slower innovation.
International reactions: who stands on which side?
Chinese restrictions have triggered a wave of reactions worldwide. Here is how key countries and organizations have responded:
United States: escalation of tensions
- The White House: In a statement on July 2, 2026, it criticized the Chinese restrictions as an "attempt to monopolize AI technology." The Biden administration announced it would accelerate work on its own export regulations for AI models, which could further intensify the rivalry between the two countries.
- Department of Commerce: Is considering adding Chinese AI firms, such as 01.AI or DeepSeek, to the Entity List, which would mean a ban on exporting American technology to these entities. This would be another step in the escalation of the tech war.
European Union: between cooperation and protection
- European Parliament: In a debate on July 7, 2026, MEPs called for a "joint EU strategy in response to Chinese restrictions." The European Commission announced a review of the AI Act regarding the protection of the open AI ecosystem.
- France and Germany: Proposed the creation of a "European open-source AI fund" worth 1 billion euros to become independent of Chinese models. This is part of a broader strategy aimed at strengthening the EU's technological autonomy.
Japan: a chance for leadership?
- Prime Minister Kishida's government announced a support package for Japanese AI companies, such as Preferred Networks or Rinna. The goal is to accelerate the development of local models and fill the gap left by Chinese projects. Could Japan become the new center of AI innovation? Sakana AI is one of the projects that could benefit from this.
International organizations: a call for cooperation
- Linux Foundation AI & Data: In an open letter on July 5, 2026, it called on China to "adhere to the principles of open cooperation" and announced the creation of a global backup repository for Chinese models. This is an attempt to save the idea of open-source in the face of growing divisions.
- IEEE: Published a report warning against the "fragmentation of AI standards," which could hinder system interoperability worldwide. This is another voice in the discussion about the future of global cooperation in the field of artificial intelligence.
Is China a pioneer? Comparison with other countries
China is not the first country to introduce restrictions on AI technology exports. Here is how their actions compare to other nations:
United States: restrictions as a geopolitical tool
- 2022–2025: Ban on the export of advanced chips (e.g., NVIDIA A100/H100) to China. This was the first major restriction that hit the Chinese AI sector.
- 2024: Export restrictions on AI models with a parameter count above 100 billion (e.g., a ban on sharing models like Llama 2 without a license).
European Union: regulations instead of bans
- 2024: The AI Act imposes obligations on "high-risk" model providers, including a requirement for audits and export restrictions to non-EU countries. This is a less restrictive approach than the Chinese one, but equally effective in controlling the flow of technology.
Russia: isolation as a strategy
- 2023: Ban on exporting AI technology to "unfriendly" countries (e.g., USA, EU) without government approval. This is part of a broader strategy of technological isolation aimed at protecting the Russian IT sector.
Comparison: China goes a step further
Chinese restrictions are more restrictive than American ones (lower parameter threshold: 10B vs 100B) and more detailed than European ones (specific criteria regarding sensitive data). This shows how seriously Beijing takes control over AI technologies.
Long-term implications: what awaits the global AI ecosystem?
Chinese restrictions are not just short-term disruptions – they have the potential to forever change the global landscape of artificial intelligence. Here are the most significant long-term consequences:
1. Fragmentation of the AI ecosystem
- Decoupling: Acceleration of the separation between the Chinese and Western AI ecosystems. Companies will have to choose between markets – e.g., Alibaba may lose clients in the US and EU but gain them in BRICS countries.
- New alliances: Countries of the Global South (India, Brazil, South Africa) could become neutral AI hubs, where models are available without restrictions. This could lead to the emergence of new innovation centers outside of China and the West.
2. Impact on national security
- Military risk: Open AI models could be used for cyberattacks or disinformation. China argues that restrictions are necessary to prevent "AI-powered hybrid warfare."
- Industrial espionage: Companies from the US and EU fear that Chinese models could contain backdoors or be used to steal intellectual property. This could lead to tighter controls on the flow of data and technology.
3. The future of open-source
- Open-source fragmentation: A possible split into "open-source with Chinese characteristics" (government-controlled) and "open open-source" (Western, decentralized). This is a threat to the idea of global cooperation that has driven innovation until now.
- Slower innovation: Restricting access to Chinese models could slow down the development of global projects, such as Meta Llama or Mistral. On the other hand, it could accelerate work on local models (e.g., the Indian Airavata or the Japanese Fugaku-LLM).
4. Economic consequences
- Costs for companies: Chinese AI startups (e.g., 01.AI, DeepSeek) may lose access to global markets, which will affect their valuations and ability to raise funding.
- Rise in AI service prices: Limiting the supply of Chinese models could raise the prices of AI services in the West (e.g., model APIs). This could hit smaller companies and startups that rely on cheap, open solutions.
What next? Scenarios for the future
The future of the global AI ecosystem depends on how other countries and companies react to Chinese restrictions. Here are three possible scenarios:
1. Escalation of the tech war
- The US and EU introduce even more restrictive export regulations, leading to a complete decoupling of the Chinese and Western AI ecosystems.
- China responds with its own limitations, e.g., a ban on importing AI models from the West.
- Consequences: Slower innovation, rising costs, technological isolation.
2. Emergence of neutral AI hubs
- Countries of the Global South (India, Brazil, South Africa) become neutral hubs where AI models are available without restrictions.
- Companies from the US, EU, and China move their projects to these countries to bypass limitations.
- Consequences: Emergence of new innovation centers, but also the risk of a "gray zone" in technology flow.
3. Return to cooperation
- International organizations (e.g., UN, IEEE) negotiate an agreement on AI model export rules.
- China eases restrictions in exchange for the lifting of American export limitations (e.g., NVIDIA chips).
- Consequences: Unlocking global cooperation, but at the cost of certain concessions from all sides.
Summary: is the era of global AI cooperation coming to an end?
Chinese restrictions on the export of open AI models are another step in the escalation of geopolitical rivalry for dominance in the technologies of the future. Their consequences extend far beyond China's borders – they affect scientific research, innovation, national security, and the economy. In the face of these changes, the key question is: is the world ready for the fragmentation of the AI ecosystem, or will it be possible to develop new rules for cooperation?
One thing is certain: the future of artificial intelligence will be shaped not only by technology but also by politics, law, and national interests. As Nature notes, this may be the beginning of a "new iron curtain" – this time in the digital world. Utopia or Judgment Day? – the answer to this question depends on how we respond to these challenges.
Comments