State media affiliated with China's CCTV broadcaster outlined the mechanics of Beijing's AI export restrictions in a late Thursday post, emphasizing that the majority of artificial intelligence technology will remain available for international distribution. The framework targets specific categories—sensitive data sources, particular use cases and computing hardware—rather than imposing blanket prohibitions.

According to the CCTV-linked WeChat account, "Since the policy was announced, two leading companies, DeepSeek and MiniMax, have successfully completed the registration process and obtained open-source and overseas licensing rights." Neither company immediately responded to inquiries about the development, though DeepSeek disclosed a $15 billion funding round from Tencent and CATL this week.

The restrictions, proposed on 16 October by China's commerce and science ministries, would add 23 AI technologies to the nation's export control list. The ministries are accepting public feedback through 15 November before finalizing the regulations.

Three Categories Under Scrutiny

The regulatory framework concentrates on three distinct domains. First, it addresses data sources: models trained using China's "core data" face outright export bans, while those relying on "important data" encounter restrictions regardless of their scale.

Second, the rules target applications deemed sensitive. Autonomous agents surpassing certain performance thresholds face restrictions, with complete bans applying to systems designed primarily for critical infrastructure, military purposes or top-level government functions.

Third, the regulations address computational infrastructure. The list includes AI chip design and high-speed interconnects for large-scale computing clusters.

The framework also establishes specific metrics for what constitutes a large model: systems requiring training compute exceeding 1026 floating-point operations, containing more than 2 trillion parameters, or ranking among the world's top ten in capability.

Commercial Channels Remain Open

Distributing restricted models via application programming interfaces, standalone applications, web interfaces, software-as-a-service platforms or lightweight on-device deployment does not qualify as technology export under the new rules. The CCTV account stated: "Simply put, AI companies can continue with their commercialization efforts as usual."

Several pathways receive automatic or expedited licensing approval:

  • Open-source releases qualify for immediate reporting and export authorization
  • Chinese companies' own data centers located outside China
  • Semiconductor manufacturing and design work conducted internationally
  • Joint research initiatives
  • Participation in international standards development

Technologies already in the public domain through open-source repositories or academic publications fall outside the regulatory scope. Exports connected to World AI Cooperation Organization member collaboration receive enhanced support.

Legal Input Shaped the Final Rules

Beijing Dacheng Law Offices attorneys Liu Jing and Liu Hongmei participated in the regulatory consultation process. They noted that certain control mechanisms shifted between the initial proposal and the published version, with many industry concerns reflected in the final text.

Liu Hongmei emphasized that companies can still move restricted technologies internationally through proper licensing channels, preventing disruption to their global expansion strategies.

Diverging From the American Model

The CCTV account positioned China's approach against Western regulatory strategies, referencing the US Commerce Department's AI diffusion framework from January 2025 and an "alliance strategy" attributed to Anthropic Chief Executive Dario Amodei. The post concluded: "But China did not take this path."

In a follow-up Friday post, the state media account reported that international internet users had encouraged other Chinese open-source AI models to pursue similar export authorization.

Source: The Next Web