It sounds right, but there is one caveat to me:

Training best models is hella expensive. Anthropic spent fortune to train it, and it definitely plans to make a fortune with it either. This US decision, if not reversed, would cause Anthropic potentially tens of billions of dollars of revenue loss. When company heads to IPO, and burning cash faster than it generates it, such moment can change their entire trajectory, plans for the future, hiring, new models development, etc.

Alright, one might say that “US will fund it directly and LLMs will move from free market to controlled and funded by government assets”.

But even then. Training new models is expensive not only in terms of computer, and not only in terms of utilities, data centers, etc. But in terms of talent either. It is hard to retain top talents with you when they should just train special models for government. I am not sure we are in 1945 again and that top tier AI researches will agree to sit in silo and work for models which only privileged selected organizations might use. Whenever government steps into control, freedom and creativity is affected.

P.S. Where I agree, though, is that we witness the start of government censorship of AI models. Imagine soon Anthropic open back access to Fable. Can we know what they put inside and which capability limitations, derived based on ID/IP, they enforce? No, we can’t. Here I agree that at least the era of government censorship begins.