That's how it is in China, officially: https://en.wikipedia.org/wiki/ICP_license
Whether it's enforced in practice (just like VPN use) is a different matter, however.
This is what the new online safety bill has you do though. It is ridiculous. And we can not even vote on it.
I think Meta released the first open model, LLAMA, because they were behind OpenAI's ChatGPT charm offensive and saw the strategic value in giving away what they had spent millions making in order to erode a competitor's leading position.
Again Deekseek gave away what they trained, and claimed it hadn't been that expensive to train anyway, in order to erode the competitor's leading position.
But with time and regulatory capture this dries up. At least within America.
There's a good chance that Deepseek basically surprised the CCP too. But they have catapulted themselves to big national and international prominence. So going forward their future releases will very much be sanctioned by the Chinese government. So we'll only see further model giveaways if it's seen as strategically beneficial to devalue the competitors in America.
(Thinking further ahead: at some point soon there might be a LLM model or augmented model that is so good at getting probabilities right that it can 'outthink' a human. At that point, whichever billionaire controls it has zero upside in sharing it with anyone. It's purpose is no longer to fuel a funding bubble as they can instead use it to outthink humans for their own personal gain. They won't think 'this can outthink humans, so we can offer better clip art pickers and other AI integrations for the masses who might pay for access to it'. They will think 'how do I use this super intelligence to outsmart the humans currently voting in democracies or humans guiding the markets?)
No one in the West would use a Chinese model otherwise, so their best strategy is to destroy the US' ability to establish dominance and more unicorns.
Expect them to keep putting out models for free.
Imagine if, to make a website, you had to fill out forms and submit them to the government for approval. Sorry, you can't have an unrestricted server connected to the internet, you can just make webpages using an approved website builder GUI using only those images that aren't filtered out by the host.
That is what the internet would be like if it were treated as the censors want to treat AI. No public training data sets. You can't run AI on your computer. You can't train your own models, or even finetune them. You can use the big censored models provided by the big companies that have the resources to get past regulatory approvals, or you can use the product of someone that has done this.
If the internet had been like this we'd be decades behind where we are today.