Keep pulling the thread on Nathan Lambert.
The White House is discussing a new executive order to manage open-source AI models.
Chinese open-source models, such as DeepSeek, currently have a substantial capability lead over other available open models.
The U.S. government is likely to ban or indefinitely delay any open-weights model with capabilities significantly exceeding GPT 5.5, Claude Opus 4.8, or GLM-5.2 within the next 6 months.
The primary driver for current AI regulation discussions is the expectation that an open-weights model will soon match the capabilities of Anthropic's Mythos model.
Anthropic is leading a political campaign against Chinese AI models by sharing blog posts and letters with government representatives.
Anthropic's advocacy for banning Chinese model makers is motivated by regulatory capture, as such a ban would provide substantial economic security for its products.
A U.S. ban on Chinese open-weight models would demolish the emerging American open-model economy, including inference and fine-tuning companies.
A unilateral U.S. ban on a frontier open-weight model would be ineffective for safety, as bad actors could still easily access and use the model if it is not also banned in countries like China.
The only way to effectively limit the progress of open-source AI is through a global agreement on managing AI model risks, which is not close to being achieved.
A potential White House executive order on open-source AI would likely impact Chinese-origin models and government use cases initially.
Open-source AI models lack a central economic champion to lobby against regulations that would negatively impact them.
Government capability thresholds for a "right to review" AI models will likely be updated much slower for open models than for closed models.