Open-source AI fight intensifies over Chinese models

Moonshot AI’s release of Kimi K3 has renewed conflict over open-source AI after the Chinese lab’s model drew comparisons to leading American systems. OpenAI’s head of strategic futures, Dean Ball, said benchmarking reportedly puts Kimi K3 on par with Anthropic’s Claude Fable and OpenAI’s GPT-5.6, but warned that open-weight models can be difficult to govern and said he was surprised China would allow models this capable to be released openly.

Ball argued that Chinese labs may be using openness because their models are behind commercial leaders and because open-weight releases create a “cloak of ungovernability.” He also described open source as “inherently decelerationist,” saying it could reduce AI capital spending and push the technology toward a state-provided public utility rather than a market product.

The debate has landed in Washington as Axios reported that the Trump administration has considered restrictions on Chinese AI labs, including placing them on the Commerce Department’s “Entity List” or making US companies responsible for risks tied to hosting Chinese models. David Sacks pushed back, accusing leading closed labs of seeking government help to eliminate open-source rivals, while also opposing broader AI regulation and state-level rules.