Why Anthropic’s Mess Doesn’t Matter & The Era of Sophistication

As you surely know, currently, any non-American like me, including immigrants living in the US, which represents a huge portion of Anthropic’s own employees, can’t use Fable/Mythos-class models by order of the United States Government (USG).

This unprecedented export control, the first on an actual AI model, has already changed AI, even if they aren’t aware of it yet, forever.

The consequences for Anthropic, OpenAI, and other suspects of receiving similar treatment are terrible.

But not because they’ve reduced their addressable market a lot if this doesn’t get overturned, which is pretty bad already, but because I believe it will start the era of sophistication, something Labs really did not expect they would have to deal with.

Let’s dive in.

The Ephimerous Fable

As I wrote about in my last issue of this newsletter, Anthropic has finally revealed its new generation model to the world.

But the release was a spectacular failure.

From sabotage to export control

Simply put, Fable was the first model you and I had no right to use freely. Depending on the use case, if Anthropic didn’t like what it saw, you would be instantly downgraded to a worse model.

This included topics such as biology, cybersecurity, or general science, areas that Anthropic considers are too dangerous to expose freely because those same topics can lead to very harmful outcomes, like cyberweapons and all that fearmongering crap you can still learn to do without Fable.

But I digress. For now, all we got was a model incapable of answering questions about mitochondria or unwilling to provide information on cancer research.

Source

Then, Dario Amodei, the CEO, published a blog talking about the huge impending risks that were coming with AI, stating, and I quote:

“Frontier AI models, like airplanes, should be required to go through technical testing and auditing, and their release should be blocked or reversed as a threat to public safety if they do not meet high standards of safety.”

And, well, to say that he got what he asked for was an understatement. A few hours later, Fable was banned.

But why?

The jailbreak that ended it all

First and foremost, let me be very clear: I have zero doubt this will be overturned soon; the fact that researchers at Anthropic, like Andrej Karpathy, can’t use the model they created is bonkers, especially given that way more than half of Anthropic’s researchers aren’t American-born or naturalized citizens.

But, anyway, the sequence, as currently reported, went as follows:

Anthropic’s version is that the government contacted it at 5:21 PM ET on June 12 with an export-control directive barring access to Fable 5 and Mythos 5 by any foreign national, inside or outside the United States, including Anthropic’s own foreign-national employees.

The government’s stated rationale was national security.

Anthropic says the letter did not provide specific details, but its understanding was that the government believed there was a way to bypass, or “jailbreak,” Fable 5’s safeguards to use it to identify cybersecurity vulnerabilities.

From all…

Enjoyed the introduction? Subscribe or sign in to continue with this edition.