The author is a Reuters Breakingviews columnist. The opinions expressed are her own. Updates to add graphic.

By Una Galani

HONG KONG, July 8 (Reuters Breakingviews) - Beijing is wielding a double-edged sword as it mulls restricting overseas access to the most advanced Chinese AI models. On the one hand, such curbs, which would include some open-source software and those yet to be released, would simply mirror severe export controls the U.S. imposed on Anthropic's Mythos, a model purported to function as ​a hacking super-weapon. What Beijing will need to consider, though, is what costs such a move would impose on its own global AI ambitions - and whether those are worthwhile.

Over the past month, at meetings led by China's Ministry of Commerce, participants discussed the possibility of putting limits on Chinese frontier models, Reuters reported on July 7, citing three people familiar with the discussions. Alibaba 9988.HK, ByteDance and Zhipu AI, formally known as Knowledge Atlas Technology 2513.HK, were among those present at the talks. Notably, the discussion included potentially restricting who can fund domestic AI startups.

Limiting access to select models may be easier and less damaging than trying to draw broad hard lines on funding. China's AI startups already operate at a big disadvantage to their U.S. peers, which share a home with the world's deepest pool of capital. In the past, unprofitable companies like $13 billion MiniMax 0100.HK and $92 billion Zhipu, whose GLM-5 product is one of the world's top-ranked open-source models, would have been considered too early stage for public markets. Their January debuts in Hong Kong's international capital market came after Sino-American tensions crushed China's U.S. dollar-led venture capital market.

More established giants like Alibaba, whose Qwen model is one of the most widely used in China and ranks highly in open-source global rankings, also rely on international capital through U.S. and Hong Kong listings. Though state-owned enterprises are stepping up as AI investors, that is not the most natural source of capital for fuelling entrepreneurial spirits.

The optics are difficult for Beijing overseas too. China's open-source models are lauded as a way to reduce dependency on U.S. ones, and support the Asian giant's soft power at a time of growing backlash against its vast trade surplus. The mere talk of restrictions from the world's two major powers will give yet more incentive to middle countries to push on with building their own sovereign-AI solutions, even if that path is more expensive. Copying some of the U.S. AI restrictions would be costly for China and the world.

Follow Una Galani on LinkedIn and X.

CONTEXT NEWS

Chinese authorities held meetings with top tech firms over the past month about potentially restricting overseas access to China's most advanced AI models, including those yet to be released, Reuters reported on July 7 citing three people familiar with the discussions.

Companies present at the talks included Alibaba, ByteDance and Knowledge Atlas, the report added. None of the three companies responded to Reuters queries.

At the meetings, led by China's Ministry of Commerce, participants discussed putting limits on the most advanced AI models — including open-source versions -- and the possibility of implementing new measures to restrict who can fund domestic AI startups, the report said.

In June, Washington ordered that foreign nationals not have access to Anthropic's most advanced Fable and Mythos models. Export controls for Fable, which is designed for the general public, have since been lifted after new safeguards were put in place. But Mythos, designed for cybersecurity professionals, is still only available to some "trusted" U.S. organisations.