China mulls AI export controls, targeting its most advanced open-weight models

China Eyes Export Controls on Its Most Advanced Open-Weight Models

China is in early-stage discussions with its top AI companies—Alibaba, ByteDance, Zhipu AI, DeepSeek, and High Flyer—about restricting overseas access to the country's most advanced models, according to Reuters. The talks remain preliminary, but they signal Beijing's growing concern that Chinese AI capabilities are leaking abroad without government oversight or benefit.

The mechanism mirrors what the US is already doing. Just as the Trump administration has restricted access to Anthropic's Fable 5 and Mythos 5 models before selectively lifting those controls, China is considering a tiered approach. At the lightest end, Beijing could require companies to file future releases with regulators or submit models for national security review before publishing weights. More aggressively, it could demand government approval before any advanced model could be released overseas. At the most restrictive extreme, China could ban public releases of frontier open-weight models entirely, limiting them instead to state-approved customers or government priorities.

The catch: you cannot claw back what's already out. Once models like Kimi are posted to Hugging Face, they can be downloaded, copied, fine-tuned, and mirrored anywhere. China would face nearly impossible enforcement challenges against open-source inference providers. The new controls would apply only to future releases.

Why now. Chinese models have become competitively dangerous. Zhipu AI's GLM 5.2 recently performed close to leading US models at much lower cost, impressing Silicon Valley. Alibaba's Qwen family has become one of the world's most important open-model ecosystems. ByteDance's Doubao is widely deployed domestically. Beijing has watched US frontier labs tighten access to their most powerful models, and it wants leverage too—and to ensure that scarce GPU resources stay focused on advancing its own frontier rather than supplying the world with cheap competition.

The broader game. China's open-sourcing strategy has been a double-edged sword. It creates pricing pressure on US providers, which hurts American labs but benefits the global AI ecosystem and tilts competitive advantage toward China's model quality. If Beijing cuts off that supply, American users and smaller companies lose access to cheap, capable models. They would be forced to either pay more for US frontier models or push other labs like Meta deeper into open source. That shift could raise costs across the board or consolidate demand back into the hands of US incumbents—which may be exactly what Beijing wants to avoid by tightening its own grip first.