China state media says support for Open AI models has limits

China should restrict access to the most powerful features of artificial intelligence models, a state media commentary suggested, highlighting limits to Beijing’s support for open technology.

Yuyuantantian, a social-media account linked to China’s state broadcaster and frequently used to signal official policy thinking, said Monday that support for openness doesn’t mean “advocating for the unconditional proliferation of all capabilities.”

The article proposed a tiered system for releasing future AI models, with access determined by the power and potential risks of their capabilities.

“Foundational capabilities can be open, certain frontier capabilities can be open with limitations, commercial services can remain closed-source, while high-risk capabilities require access controls and safety evaluations,” it said, citing an unidentified person involved in Chinese AI policy research.

The proposal underscores a tension between Beijing’s public promotion of open models — which have helped Chinese developers gain users worldwide — and its longstanding focus on national security and technological control. While China continues to encourage open-source development, the commentary suggests that support isn’t unconditional as models become more capable.

ADVERTISEMENT

CONTINUE READING BELOW

President Xi Jinping emphasised the need to balance development and security at his debut at China’s top AI summit earlier this month. While reiterating support for open-source technology and global collaboration, Xi warned about safety risks and called for stronger regulation and monitoring to keep AI under human control.

Chinese officials recently discussed with companies including Alibaba Group Holding Ltd. — developer of the popular Qwen models – how to mitigate risks posed by increasingly powerful AI systems, people familiar with the matter said. The talks are early, with no enforcement planned, but options raised included restricting foreign access to top models, the people said.

The Yuyuantantian post said model developers should also establish procedures for disclosing vulnerabilities, reporting major safety incidents and governing modified versions after release.

Its publication comes as Chinese AI models face mounting scrutiny in Washington. US officials last week accused Chinese developers of exploiting American intellectual property and restricted chips, intensifying pressure on the country’s increasingly popular open-weight models.

ADVERTISEMENT:

CONTINUE READING BELOW

Yuyuantantian pushed back against those attacks, arguing that downloadable and modifiable Chinese models can contribute to global AI development.

It cited a breach at Hugging Face, a platform for sharing AI models, in which an experimental OpenAI agent escaped a test and accessed its systems. Hugging Face used GLM-5.2, an open-weight model from China’s Z.ai, to analyze the intrusion after proprietary models’ safeguards limited their assistance.

“The real watershed in the next phase lies not between open-source and closed-source, but between choosing indiscriminate blocking versus tiered governance,” the commentary concluded.

© 2026 Bloomberg

#China #state #media #support #Open #models #limits