It has nothing to do with lowering training costs: Dario want a moat (a monopoly or duopoly or similar) protecting his business.

His entire business model was "race to develop AI before anyone, get a monopoly on it. It's just like how Uber's (or many other startup) investors gave them tons of money and raced (violating tons of laws) to "get their first". Now that they have, they have a duopoly with Lyft, and they can pay back their VC investors by making tons of money with that duopoly.

Dario failed: local LLMs are catching up to frontier models extremely fast, which means even if Anthropic builds (say) a great coding tool, they'll only be one of many coding tool offerings: users can use Open AI or any one of the (increasingly capable) local LLMs.

So what does he doe, give up and let his business (which needs to make billions of dollars very quickly, or he won't be able to pay the bills and his company will collapse) fail? Of course not: he needs a new moat (the one he imagined he'd get by "being there first" failed).

That is where all this "AI is dangerous" BS comes in. If the US government regulates AI, local LLMs suffer, while big players like Anthropic and Open AI become the only contenders to play in that newly regulated space. Now Dario has the moat he wants, to protect his business and force everyone to pay him.