Lambert's argument, published on his widely-read Interconnects newsletter and among the most-shared AI reads of the weekend, is that two policy debates — distillation of US models by Chinese labs, and frontier open-weight capabilities — are converging into a platform of support for restricting open models. Multiple sources cite White House discussions of a new executive order. The most likely action, he writes, is a ban or indefinite delay on open-weight releases meaningfully above the current frontier tier, which with today's capability gap would almost certainly hit Chinese-origin models first.
His deeper concern is structural: open models lack an economic champion. Closed-model companies have far more effective lobbying, and Lambert describes the anti-Chinese-model campaign — led most visibly by Anthropic's policy letters — as having crossed from genuine business concern into regulatory capture, since a ban on the accused Chinese model makers would directly benefit the companies pushing for one. At a June 9 meeting on the government's model-licensing framework, a representative of US open-source lab Reflection AI argued for capability-based exemptions.
The counterpoint arrived from Beijing the same weekend. Zhipu founder Tang Jie published a memo arguing frontier AI must remain open, and the company shipped GLM-5.2 — a model Lambert himself places at the frontier tier — free to download and commercialize. The gap between the two positions is now the sharpest fault line in AI policy: one side moving toward capability ceilings on open weights, the other making frontier capability a free download.