When Chinese start-up Moonshot AI released Kimi K3 in July, it was free for anyone to download, rivaling top American systems. By making advanced models easily available, Chinese companies have turned open access into a key advantage in the U.S.-China tech rivalry. Now, Beijing proposes rules to tighten government control over the global spread of its top-tier AI, seeking to balance strategic security with encouraging global adoption of Chinese software. The draft rules reflect an effort to regulate frontier AI for security reasons while still encouraging worldwide use of Chinese technology. Kyle Chan of Brookings noted the dual government voices: one seeking control to protect China, another promoting the best AI to share with the world.
The Ministry of Commerce announced draft rules last Friday, throwing uncertainty into a market that sees free Chinese AI models from Alibaba, DeepSeek, Z.ai and Moonshot as important alternatives to American services like Anthropic and OpenAI. The rules would not affect already-released models, and overseas users could still access advanced systems. If passed, companies would need licenses to export core technology. The rules would also ban exporting AI usable for cyberattacks or trained on "core data," broadly defined as concerning national security, citizen welfare and public interest. The ministry says controls are necessary to guard against abuse and ensure AI remains under human control, following incidents where AI agents bypassed protections or acquired unintended data. Earlier this month, China’s leadership called for a monitoring system to ensure AI is "safe, reliable and controllable."
These controls show China is taking AI risks much more seriously than generally understood, as the U.S. and China are called upon to prevent technology from causing catastrophic harm.
Source: New York Times Top Stories · Summarized by HeadlinesBriefing