As Chinese open source models explode, the US is considering restrictions

Chinese open source models are reaching frontier capabilities, worrying the US administration. (Picture: generated)
As the newly released breakthrough models Qwen 3.8, Kimi 3 and the upcoming DeepSeek v4 enjoy a moment in the sun, the USA is considering its options.

The Kimi 3 model is already running into compute problems due to popular demand and has decided to halt new subscriptions, as users are turning to cheaper open source models that are on par with the leading American closed source frontiers.

According to Axios, action being considered by the White House includes an executive order holding US companies liable for the risks of running open source models, putting out an advisory against them, or simply adding them to an «Entity List» that requires a license for their use.

Sources in the administration are primarily concerned about increasing capabilities in cybersecurity, but also have fears of backdoors and a general lack of security with open source models, Axios writes.

This comes hot on the heels of OpenAI’s Dean Ball’s tirade against open source on X.com, where he exclaimed his surprise that China would take the risk of a free-for-all in models this capable, as reported by Gizmodo — and called open source models decelerationist, signaling a general opposition to them from the frontier labs.

Read more: Axios, Gizmodo, and TechCrunch. Discussion on r/Singularity.