DeepSeek and MiniMax Cleared to Export AI Models Abroad
Chinese AI leaders DeepSeek and MiniMax have become the first firms cleared to export their AI models abroad under new, clarified national export controls. While the regulations target sensitive data, critical applications, and advanced computing power, they largely exempt commercial deployments like APIs and open-source technologies. This move signals China's intent to balance national security with global AI commercialization.
Leading Chinese AI firms DeepSeek and MiniMax have secured crucial licensing to export their artificial intelligence models for overseas use, according to a report from Yuyuan Tantian, a WeChat account linked to state broadcaster CCTV. This development marks a significant step under China's newly clarified AI export control framework, which aims to regulate sensitive technologies while largely permitting broader commercialization globally.
The announcement, made in a post late on Thursday, indicates these companies are the first to successfully complete the registration process. DeepSeek and MiniMax did not immediately respond to requests for comment regarding the report, which also noted DeepSeek's recent reported fundraising of $15 billion from Tencent and CATL.
China's Targeted AI Export Controls Take Shape
The move follows draft AI export controls proposed by China's commerce and science ministries on October 16. These regulations, open for public comment until November 15, seek to add 23 AI technologies—including large models, agents, and computing clusters—to the nation's restricted export list. Companies must register before exporting any listed technology.
The CCTV-affiliated account detailed that the controls primarily target three critical areas:
- Data: Exports of large models trained on China's “core data” are banned, while models using “important data” face restrictions regardless of their size.
- Sensitive Use: AI agents exceeding specified performance levels or those built for critical information infrastructure, defense, or core government affairs are either restricted or banned.
- Computing Power: The regulations also list AI chip design and high-speed links essential for large computing clusters as restricted items.
Specific quantitative thresholds for large models under these rules include training compute exceeding 10^26 floating-point operations, more than 2 trillion parameters, or a capability ranking in the global top ten.
Commercialization Efforts Largely Unaffected
Crucially, the new framework clarifies that many common commercial applications of AI will remain outside the scope of these controls. The post explicitly states that offering restricted models via API calls, applications, web pages, Software as a Service (SaaS), or lightweight local deployment does not constitute a technology export.
“Simply put, AI companies can continue with their commercialization efforts as usual,” the post affirmed. Furthermore, open-source technology is granted automatic licensing, described as “immediate reporting and export.” Other exemptions include Chinese companies' own computing centers abroad, overseas chip tape-out, joint research initiatives, and participation in international standards development. Technologies already made public through open source or academic papers are also exempt.
Legal experts from Beijing Dacheng Law Offices, Liu Jing and Liu Hongmei, who participated in the consultations, noted that several company concerns raised during the first draft phase have been incorporated into the published version. Liu Hongmei emphasized that restricted technologies can still be exported with a license, ensuring that companies' overseas plans are not disrupted.
A Distinct Approach to AI Governance
China's clarified stance contrasts with the US approach, which the post referenced by citing the Commerce Department's AI diffusion framework of January 2025 and an “alliance strategy” attributed to Anthropic CEO Dario Amodei. The state media account asserted that “China did not take this path,” suggesting a different philosophy in managing AI technology's global dissemination.
This early approval for DeepSeek and MiniMax, two prominent players in China's burgeoning AI sector, signals Beijing's intent to foster global participation for its tech giants while carefully safeguarding national interests in critical AI capabilities. The policy aims to strike a balance, allowing the bulk of commercial AI innovation to flourish internationally under a more transparent regulatory environment.
FAQ
Q: What types of AI technology are restricted under China's new export controls?
A: The controls target large models trained on China's “core” or “important” data, AI agents for sensitive uses like critical infrastructure or defense, and computing power related to AI chip design and high-speed links for large clusters. Specific thresholds for large models also apply.
Q: How do Chinese AI companies like DeepSeek and MiniMax benefit from these clarifications?
A: DeepSeek and MiniMax are among the first to receive licenses, allowing them to export their AI models overseas. The clarifications also ensure that commercial activities like offering models via APIs, apps, SaaS, and open-source deployments are largely exempt from strict export controls, facilitating global market access.
Q: Does this policy mean all Chinese AI models can be freely exported?
A: No, not all models can be freely exported. The policy targets specific, sensitive AI technologies and applications. However, it explicitly exempts most commercial deployments (e.g., via APIs or SaaS) and open-source technologies, meaning a significant portion of AI innovation remains free for international use with proper registration or automatic licensing.
Related articles
Here’s where you can preorder the new OLED-equipped iPad Mini: Apple
Apple has officially opened preorders for its eighth-generation iPad Mini, featuring a vibrant 8.5-inch OLED display and powerful A20 Pro chip. Starting at $699, the device can be preordered from Apple and Best Buy before its November 6th release, promising significant upgrades in visual technology and performance.
Xreal Unveils Aura Wired Smart Glasses Running Android XR
At the Android Dev Summit on Wednesday, Xreal, Google's smart glasses collaborator, officially introduced its new Xreal Aura wired smart glasses. The device, which had been teased earlier in the year at Google's I/O
ICE's Covert Anti-Left Surveillance Operations Uncovered
New evidence reveals ICE is running a vast, potentially unconstitutional surveillance operation against left-wing activists nationwide. Leveraging post-9/11 tools and extensive data, experts liken it to a 'COINTELPRO 2.0,' marking an unprecedented scale of domestic spying.
Sakurai Shrugs Off AI Games: Why Human Creativity Still Reigns Supreme
Masahiro Sakurai, creator of *Super Smash Bros.*, believes most AI-created games will be "unprofitable titles" and "low-quality goods." He argues AI can only reproduce existing systems, not generate truly original artistic works like those from Fumito Ueda. In an era of mass production and industry shifts to AI-first models, Sakurai stresses the growing importance of unique human creativity and vision.
GM Pivots to Hybrids with $2 Billion Investment in Lansing Plant
In a significant strategic shift, General Motors has announced plans to invest over $2 billion in its Lansing region operations, including $1.25 billion specifically for its Lansing Delta Township Assembly plant. This
The Super Mouse puts a touchscreen macro pad under your hand: Tech
Arctic Fox Technologies has launched the Super Mouse, an innovative peripheral featuring a 3.1-inch touchscreen for dynamic app controls, macros, and enhanced productivity. Available on Kickstarter for $199, this wired mouse aims to consolidate multiple desktop devices into one intuitive command center. Its contextual display and advanced features redefine user interaction.