WTF is going on?!

I cannot believe what just happened.

Fable and Mythos both gone. The US

government citing national security

authorities has issued an export control

directive to suspend all access to Fable

5 and Mythos 5 by any foreign national

and I'm going to explain exactly what

that means. whether inside or outside

the United States, including foreign

national anthropic employees. The net

effect of this order is that we must

abruptly disable Fable 5 and Mythos 5

for all our customers to ensure

compliance. Access to all other cloud

models is not affected. We apologize for

this disruption to our customers. We

believe this is a misunderstanding and

are working to restore access as soon as

possible. And I'm going to get into the

full blog in a moment, but there's a few

things that I want to point out here.

First, what is a foreign national? It's

basically any non US citizen, which is

crazy because there are obviously a ton

of green card holders, just as an

example, in the United States that work

for these large tech companies,

including people like Andre Karpathy,

one of the greatest minds in artificial

intelligence who just about two weeks

ago joined Anthropic. And so, how is

Anthropic supposed to enforce this rule?

How are they supposed to know whether

someone's a citizen or a non-citizen

when they're just using the API? They're

going to have to do things like know

your customer, which is what financial

institutions have to do. They basically

have to gather all of this information

about their customers to help report on

different financial crimes. Now, are AI

companies going to have to do that? The

net effect of this order is we must

abruptly disable Fable 5 and Mythos 5.

Within a minute of this tweet dropping,

Fable 5 and Mythos 5 stopped working. I

literally had 10 different agents

running on Fable 5 right when this tweet

went out and they all just stopped. But

let me tell you something. This is a

self-inflicted wound on Anthropic. Now,

some of you might think that this is

actually a big marketing W. And I'm

going to get to that in a moment. But

from my perspective, this is a big

problem. Anthropic is now serving

millions of customers with Fable 5. And

all of a sudden, all of those customers

have to scramble and figure out another

model to use. But I said this is

self-inflicted. Why? It all started with

this project Glasswing. This is the

first time that we heard about Mythos.

This was a few months ago. And anthropic

in the most anthropic way said, "Hey, we

have this new training run. We have this

new incredible frontier model called

Mythos, and it is so good, we can't even

release it. It is too good at cyber

attacks, cyber defense, and you can't

use it. We can use it, by the way. You

know, we can use it. We got it. We're

trustworthy enough. We're the good guys

here. So, we're going to use it, but

we're not going to release it." Well,

actually, maybe we'll just choose a

handful of other companies that we deem

good enough to use this model, but none

of you others can use it. And this is

fear-based marketing. And because they

took this tactic of saying it's too

dangerous to release, everybody wanted

to use it. Everybody was talking about

it. And then just before they talked

about Project Glass Wing, they also had

their little scuffle with the Department

of War. Anthropic put a line in the sand

saying, "No, the Department of War

cannot use our models for fully

autonomous weapons and to do mass

surveillance on US citizens." And the

Department of War said, "No, we don't

want you to tell us how to use your

models. We're going to follow the letter

of the law." And so the government

deemed anthropic a supply chain risk.

This was a very big deal. The first time

a US company had been deemed a supply

chain risk. So all of this was happening

and then fast forward to just last week,

Anthropic launches Fable 5, which is

basically just mythos, but with a

classification model on top that

prevents people from asking about cyber

security questions, bio questions,

nuclear questions. It kind of is a

neutered version of Mythos. It's still

incredibly powerful. Don't get me wrong,

it's the best model I've ever used, but

again, this is more of that fear-based

marketing. It's, hey, we have this

mythos version we can't give you, but we

we'll give you this one over here. And

then that brings us to now the US

government told anthropic non- US

citizens can't use it. Of course, if

they talk up how dangerous this model

is, how only they are good enough to use

it and good enough to deem others who

can use it, of course the government is

going to look at it and say, "Well,

maybe non- US citizens shouldn't use

it." Okay, and here is where the story

gets wild. It turns out the government

decided to make this decision because

they found out that Fable can be

jailbroken, which is kind of like

finding out the sky is blue because

every single model ever can be

jailbroken. Obviously, the rate of being

able to jailbreak these models continues

to decline, but they are humanlike in

that sense. They can be convinced to say

things that they shouldn't. That is the

nature of artificial intelligence in its

current form. That's why I was a little

bit surprised to see that the government

found out that these models are

jailbreakable because all models are

jailbreakable. And there's nothing

different about Mythos. There's nothing

different about Fable. But what's crazy

is that the government did find out. And

apparently it was an Amazon employee who

told them. Listen to this. This is from

the Wall Street Journal. The jailbreak

research in question was done by

researchers at Amazon who used a series

of prompts to get Anthropic's model to

provide them with information about a

handful of security vulnerabilities,

said chief executive with the cyber

security firm Luda Security. Anthropic

shared a copy of the report with her.

she said. And breaking just as of a few

minutes ago, the information reports

that the Amazon CEO himself, Andy Jasse,

was amongst the tech leaders to push the

Trump administration to consider the

severe security risks of Mythos and

Fable. So, according to the information,

Amazon CEO Andy Jasse was among the tech

leaders who raised concerns to senior

Trump administration officials this week

about security risks in Anthropic's most

advanced models. According to two people

familiar with conversations, the calls

between the head of Amazon, one of

Anthropic's biggest investors and

vendors and the officials in the last

few days helped set in motion the Trump

administration's new export restrictions

on Anthropics Claude Mythos 5 and Fable

5 models late Friday night. This is

unbelievable.

So, has nobody heard of Ply the

Prompter? I mean, he literally jailbroke

it within hours of the model being

released. Here it is. Ply the Liberator

jailbreak alert. Fable 5 liberated. I

mean, this is literally what he does. He

jailbreaks every single model that comes

out within hours of it and generally

within minutes. All right, now let me go

through the full response blog post from

Anthropic about the government banning

non- US citizens from using these

models. So, one, they received the

directive from the government at 5:21

p.m. Eastern. So within 3 hours of

receiving that directive, they shut down

the model, which is kind of insane to

think about. Imagine all of the

companies that are so dependent on the

model and they just flip a switch and

that's it. They no longer have access.

The letter did not provide specific

details of its national security

concern. Well, it's possible that these

things can be jailbroken. You told us

how dangerous they are. So why would we

want foreign nationals using it? And

they told him, listen to this. The

government believes it has become aware

The government believes it has become

aware of a method of bypassing or

jailbreaking Fable 5. We reviewed the

demonstration of this specific technique

being used to identify a small number of

previously known minor vulnerabilities.

These vulnerabilities all appear

relatively simple and we have found that

other publicly available models are able

to discover them as well without

requiring a bypass. So there's a few

things here. One, again, all models are

vulnerable to jailbreaking. There's no

preventing it. There's just minimizing

it. Two, it's fairly straightforward to

get the models to talk about bioweapons,

to talk about nuclear reactors, to talk

about things that they really shouldn't.

And that kind of information is

generally available online anyways. But

where it actually does become dangerous

is when the model is so good at cyber

capabilities that it's able to discover

vulnerabilities in software that nobody

else knows about. But here's the thing,

Mythos Fable, they're not that much

better than GPT 5.5 at doing so. But of

course, again, the self-inflicted nature

of this wound for anthropic, all of the

fear-mongering that they have done

around how dangerous mythos can be in

the wrong hands, they are getting a

taste of their own medicine now. And

guess who's not? Open AI. Open AAI is

still serving a model that is nearly as

good. And I guarantee GPT 5.6 is going

to be even better at these types of

capabilities. and they're going to be

jailbreakable just like every other

model on the planet. So, here's some

other information that we found out from

this blog post. Obviously, Anthropic

have instituted strong safeguards that

greatly reduce the likelihood that Fable

is misused. However, it's not perfect.

It can never be perfect. That is the

nature of these models. In the weeks

leading up to the launch of Fable,

Anthropic worked with the US government,

the UK, AISI, multiple private

thirdparty organizations, and internal

teams to red team Fable safeguards for

thousands of hours in total. I wonder if

they worked with Ply. I don't think they

did, and they really should have because

he can jailbreak any of these models.

There is no preventing Ply from plying.

And these tests show that fable

safeguards are substantially more

effective than those of any previously

deployed model. But here's the thing. If

the capabilities of this model are so

much better than every other model, that

negates the fact that the safeguards are

also better. It only takes one time to

get a critical vulnerability out of

Fable to have massive damage on the

United States. No testers have yet been

able to find a universal jailbreak,

which means a single jailbreak that just

works every single time. That doesn't

exist, and that's okay. It's not really

necessary to exist. Again, you only

really need to get it to reveal

sensitive information once. Now, a lot

of people are also saying that this is a

huge W in terms of marketing. And yeah,

I guess in a way it is. A lot of people

are going to read about this. In fact,

this post by Ananthropic already has 66

million views in less than 12 hours. So,

here's Ryan Peterson, the CEO of

Flexport. Such great marketing reminds

me of when this shoe company got the NBA

to ban their shoes for making you jump

too high. That's exactly what this is,

right? A lot of people are going to find

out about it. A lot of people are going

to say, "Wow, it was such a good AI

model that the government had to ban it.

I got to try that out. Why would I use

OpenAI as non-banned models? I mean,

they can't be as good. They're not

banned. And I actually think that makes

a lot of sense. But I do believe the

disruption to Anthropic's business is

going to be far worse than any marketing

they receive. Remember, Anthropic just

confidentially submitted their S1,

basically setting themselves up to go

public. This is almost certainly going

to delay that because now they are not

selling a product. They are now selling

a national security threat. This is much

different. This changes the calculus

completely. And of course, the SEC is a

government organization, so they have to

take this into consideration. Investors

have to take this into consideration.

So, here's the Department of War's CIO,

Chief Information Officer, Kristen

Davies. We fully support POTUS and the

Secretary of War and prioritizing

national security and the security of

our war fighters, DIIB partners,

critical infrastructure, international

parties, and allies. Here's the most

important part. Some things are simply

more important than revenue cycles,

clickbait, and pre-IPO valuation. She is

specifically saying anthropic, you did

this to yourself. By the way, I don't

necessarily agree with the ban. Just to

be clear about that, I think finding a

very narrow vulnerability, a very narrow

jailbreak in these models is not enough

to warrant banning them. And also the

broad brush in which they did it. any

non- US citizen can't use it. There is

truly no effective way to enforce that.

So, what happens next? Well, almost

certainly their IPO is going to be

pushed back. What we're probably going

to see is in the next 1 to two weeks,

they're going to work out a deal with

the government and we're going to get

Fable 5 back. I'm willing to bet on

that. This is the tried andrue Trump

administration playbook. Make some

extreme decision, then negotiate and

come back somewhere in the middle.

That's what they do. They might require

Anthropic to more aggressively collect

information about their customers, which

would be a major pain for Anthropic. But

at the same time, Anthropic might

actually be down with it because that's

easy for them to do when they're a

trillion dollar company. They have all

the funding in the world. They have a

massive employee base, but that's

actually going to hurt startups looking

to compete with them in the future.

That's also known as regulatory capture.

So maybe Anthropic and Daario are

playing a little bit of 4D chess here

and they kind of knew the government was

going to do this and they knew that they

were going to tighten restrictions. In

fact, Anthropic is kind of known for

being super aggressive with regulatory

capture. In fact, just a couple days

ago, Daario released an essay called

Policy on the AI Exponential, basically

saying the government is not keeping up

with how fast AI is evolving. and they

laid out all of these regulatory

frameworks that they thought the US

government should employ to safeguard

against AI development and this is

regulatory capture. They literally said,

"Hey, do this to us." And the government

was like, "All right, we'll do it." And

then and then Daario is, "Whoa, whoa,

what happened there? We don't agree with

this." And so this actually marks an

important point in the timeline of

artificial intelligence. We are now at

the point at which the government and

the general population looks at these

models as national security threats, not

as tools. This is a really bad thing

because now every model after this is

going to be viewed through the same

lens. How is this going to hurt the US?

How is it going to hurt our allies? Not

how is this going to help humanity? How

is it going to increase productivity,

increase GDP, increase the wealth of the

world? The conversation was already

extremely negative. The perception of AI

was already extremely negative,

especially in the United States. And

this makes things so much worse. Now,

I've made multiple videos about

anthropics fear-based marketing. Here's

one right here you should check out

because it's extremely relevant to this

topic.