So what’s the truth? Why did the US government block Anthropic’s Fable and Mythos models? What’s going on behind this whole soap opera?
It’s complicated, as you know, and in every moment of genuine existential crisis in my life I come here to Lidl to shoot my videos. And I want to be honest with you: I haven’t fully figured out what’s happening yet, I’m still doing my homework. But before I try to give you a more complete picture, I want to point out a couple of details - maybe you can help me make sense of them too.
It’s not a full ban - it’s a restriction by nationality
Here’s a detail that matters a lot: they didn’t take the model down for everyone. The requirement is that it can only be used by American nationals - it must not be available to foreigners. Nobody said “shut it down”; Anthropic then decided to pull it down on its own.
And the report, interestingly, came from Amazon - from Amazon researchers. Amazon invests in Anthropic; it’s one of its biggest investors.
Now, the US government doesn’t have any particular interest in tearing down half of the AI industry, because a large part of GDP comes from there. American GDP growth comes above all from the AI sector. So why would they want to break it?
The guardrail story
Then there’s this story about guardrails - that Amazon researchers managed to break through the guardrails, and therefore the model is supposedly too dangerous.
Here’s how I read it. Those guardrails, those classifiers that judge how dangerous what humans ask the model is - Anthropic’s researchers are very good at building them, they’ve been doing it for ten years. They’re made deliberately sensitive, they trip immediately. And this connects to the assumption behind everything I’ve been saying for the last two or three years:
The fundamental key to understanding all this madness is that compute is limited. They don’t have the infrastructure to give this stuff to everyone. So “it’s dangerous, let’s only give it to a restricted group” becomes a convenient framing - these people get the premium tier. In other words, I don’t think it’s really about danger.
What worries me is the State stepping in
But now that the State enters the picture and tells you that nationality matters - that it hinges on the nationality of whoever uses this stuff - that points me in a very specific direction. It reminds me of what happened in the UK with the block on pornography.
To me, these maneuvers - especially once the State gets involved - smell like surveillance. This kind of thing requires an ID document. So it’s not really about the foreigner versus the national. It’s a way to start quietly asking for documents, like a bank does: give me your ID, I have to check who you are, everything has to be collected, filed, monitored.
And this goes hand in hand with the push toward central bank digital currencies - these digital coins that ape and copy Ethereum and crypto, which have made a big leap in the conversation among administrators. In China it’s already happened; in Europe it’s being discussed; in the US too. These are all ways of tying the digital to identity, and identity to a digital currency.
I know it sounds like a conspiracy theory. I can put on the tinfoil hat. But this stuff points way too far in that direction. Because - seriously - what does it matter, if you use artificial intelligence, whether you’re American or not?
The block on open models
The other thing that stinks here is the block on open models, particularly the Chinese ones. What better narrative to:
- Start asking for documents from anyone who uses AI, and
- Create a regulation whereby models must be closed and must go through a series of certifications from the administration in order to be used.
In fact, the latest blog post that [Dario] Amodei wrote says exactly this: we need a certification process for models, similar to the one used for airplanes.
On one thing I agree with the American state, with Anthropic, and with the big digital players: the Chinese are eating them alive. So what better occasion than the “extremely dangerous model” to say: one, if you use it, give me your ID, because I need to know who you are, what you’re asking, where you are, what you do, what you want to do - I need to profile you to death, like they’ve always done. And two, start putting in place regulations that give American big tech a blank check while blocking what comes in from China.
My suspicion
I have the suspicion it’s something along these lines: digital identity, a block on Chinese models, and maybe this currency angle mixed in too - digital money - because I see signals of this pretty much everywhere.
I might be wearing the paper hat, but that’s where all of this is pointing.
Anyway - take care. Today at 7:00 there’s a live with Giuseppe Funicello. We’ll talk about what role building digital products still plays. Is it still worth making digital products? Is it still worth creating and using frameworks now that we have vibe coding? We’ll talk about it live - it’ll be interesting, human, philosophical. We’ll talk heart to heart about our future.
One last PS: if you’re from Lidl and you’re watching me, I’m your guy. The only sponsor I’d ever accept in life is Lidl. I’ve had plenty of offers, even from the famous big tech companies, and I’ve told them all to get lost. Lidl, though - Lidl I’d accept, I’m telling you.