China's AI Export Restrictions
China Considers Limiting Overseas Access to Advanced AI Models Amid National Security Concerns
Key Points
Three-tier release framework proposed by a May Supreme People’s Court roundtable, outlining: basic open-source tools would require a government filing; more advanced technologies would undergo security reviews; and the most sensitive frontier models could be barred from public release or restricted to domestic use.
U.S. actions that parallel export controls have targeted large language models, with reports that Anthropic’s Mythos and OpenAI’s GPT-5.6 faced restrictions over safety concerns.
Z.ai released GLM 5.2 last month, claiming it matched Mythos’ ability to spot bugs, underscoring the ongoing race to prove safety and reliability of frontier AI models.
Open-weight releases have driven global adoption by Chinese models, with weights available for download and used by some U.S. firms to cut costs; restricting access could disrupt this ecosystem and raise costs for foreign users.
China is considering blocking foreign access to its most advanced AI models, including ones not yet released, Reuters reported on July 7. China's Ministry of Commerce held closed-door meetings with Alibaba, ByteDance, and Z.ai to draft the restrictions — a sharp reversal from Beijing's earlier strategy of flooding the world with cheap, high-performance open AI tools.
The proposals would cover both closed-source and open-weight models — meaning even freely downloadable AI weights could be locked inside China. Leaking or stealing AI technology could become a national security crime. The scope is not final, and no timeline has been set, Decrypt reported.
Three-Tier Framework Would Sort Models by Risk Level
A May 2026 legal roundtable published in a Chinese Supreme People's Court journal proposed a three-tier system. Basic open-source tools would need only a simple government filing. More advanced models would require full security reviews. The most sensitive frontier models could be banned from public release or locked to domestic use only.
The framework mirrors how Beijing already handles nuclear and aerospace technology — treating them as strategic assets rather than trade goods. Officials also floated limits on foreign investors funding domestic AI startups, which would cut off a key source of outside capital, according to Reuters.
Open-Weight AI Has Quietly Reshaped Global Costs
Chinese open-weight models exploded in global use after DeepSeek's R1 launched in January 2025. On OpenRouter, a popular model hub, Chinese models went from under 2% of token usage in late 2024 to roughly 61% by mid-2026. Many U.S. and European firms switched to models like Alibaba's Qwen and Z.ai's GLM to cut API costs.
Z.ai released GLM-5.2 last month under a free MIT license. It scored 51 on the Artificial Analysis Intelligence Index — the highest of any open-weight model — and matched top U.S. systems with a 1-million-token context window. Restricting these models would force foreign firms back onto pricier Western alternatives, analysts warn, according to Crypto Briefing.
China Is Copying the US Playbook on AI Controls
The move follows direct U.S. government interventions in June 2026. The Commerce Department ordered Anthropic to cut off foreign access to its Mythos model over national security concerns. OpenAI then gated its next-generation GPT-5.6 to just 20 partner organizations, bypassing a planned public launch, Head Topics reported.
Scott Singer of the Carnegie Endowment for International Peace said China faces the same hard trade-off Washington does: "China is going to have to balance the benefits of access to global markets with a desire to control a highly unpredictable technology." Both countries now treat frontier AI the way they treat weapons — as something too dangerous to share freely, according to Decrypt.
Europe and Global Developers Face the Biggest Disruption
Europe is especially exposed. Its €200 billion InvestAI initiative is behind schedule and outspent by U.S. tech giants. European companies have leaned on cheap Chinese open-weight models to avoid dependence on expensive American APIs. Losing that option would squeeze budgets and push firms toward a handful of pricey Western providers.
The global developer community warns an AI "haves and have-nots" divide is forming. As Crypto Briefing noted, both Washington and Beijing are now moving in lockstep — each citing safety risks from dual-use AI capable of cyberattacks and biological agent design. The era of freely shared frontier AI may be ending faster than most expected.
Sentiment
Mostly NegativePublishers
31
Articles
136
Reach
167
