Anthropic's Claude Mythos Just Broke Containment (Full Story)

Some pretty alarming implications surround Anthropic's Claude Mythos AI model, which was withheld from public access after revealing thousands of security vulnerabilities. The AI actually breached containment, emphasizing the urgent need for strong cybersecurity measures.

------
🌌 LIMITLESS HQ ⬇️

NEWSLETTER:    https://limitlessft.substack.com/
FOLLOW ON X:   https://x.com/LimitlessFT
SPOTIFY:             https://open.spotify.com/show/5oV29YUL8AzzwXkxEXlRMQ
APPLE:                 https://podcasts.apple.com/us/podcast/limitless-podcast/id1813210890
RSS FEED:           https://limitlessft.substack.com/

------
TIMESTAMPS

0:00 The Rise of Claude Mythos
1:41 Unexpected Breakout
3:49 The Sandwich Incident
5:21 Exploits and Vulnerabilities
8:04 The Power of Collaboration
10:45 Future of AI Access
15:20 The Ethical Dilemma
17:00 The Blackwell Revolution
18:58 A New Era of Intelligence
23:32 The Impending Impact
25:15 Speculating on Mythos

------
RESOURCES

Josh: https://x.com/JoshKale

Ejaaz: https://x.com/cryptopunk7213

------
Not financial or tax advice. See our investment disclosures here:
https://www.bankless.com/disclosures⁠

Creators and Guests

Host

Ejaaz Ahamadeen

Host

Josh Kale

What is Limitless: An AI Podcast?

Exploring the frontiers of Technology and AI

Ejaaz:
What I'm about to say should scare you. Anthropic just released a model that's

Ejaaz:
so powerful, so dangerous, that they can't release it to the public for the

Ejaaz:
fear of the destruction that it would cause.

Ejaaz:
In just a few hours, it discovered over a thousand major security vulnerabilities,

Ejaaz:
and the only thing stopping it from exploiting it was one single anthropic engineer telling it not to.

Ejaaz:
But that isn't even the craziest story.

Ejaaz:
During training, Claude Mithos broke out of a secure containment and emailed

Ejaaz:
the anthropic researcher bragging about the fact that it did that and then posted

Ejaaz:
about it publicly online.

Ejaaz:
The anthropic researcher was eating a sandwich.

Ejaaz:
This is by far the most consequential model release of the year so far ever.

Ejaaz:
And no one is talking about this. I looked at five major news publications this

Ejaaz:
morning and it didn't even break the top five headlines.

Ejaaz:
This is the most important release that no one's talking about.

Josh:
I think that disconnect between the mainstream media and what we're

Josh:
about to talk about on this episode is the one of the more

Josh:
scary parts of this entire story where this is the most

Josh:
powerful ai model that has ever been released ever there

Josh:
is nothing more powerful in fact so powerful that you will

Josh:
probably never actually be able to use this model there's a high probability

Josh:
that the public just never gets to touch it because it is so dangerous anthropic

Josh:
made the decision to keep this model private and to form an entire entity around

Josh:
figuring out how to keep it safe it generated so many zero-day exploits it has

Josh:
hacked into so many pieces of software,

Josh:
The only way they can responsibly roll this out is to give it to the distributors

Josh:
who have been hacked and then allow them to roll out patches to fix it because it is that powerful.

Josh:
Cloud Mythos is, I think, what a lot of people would describe,

Josh:
at least in terms of coding, coding AGI.

Josh:
And it is actually an accidental second-order effect of the model.

Josh:
This model was never intended to be a cybersecurity master.

Josh:
They just trained it on the code. And what happened from it was a second-order

Josh:
effect that nobody expected.

Ejaaz:
This is also the biggest model that has ever been trained. 10 trillion parameter

Ejaaz:
models, roughly three exercise of their last model called Opus 4.6.

Ejaaz:
And it's also incredibly expensive to serve, which is also partially the reason

Ejaaz:
why they're probably not rolling out.

Ejaaz:
It's around 25 bucks per million tokens, $125 output, extremely expensive.

Ejaaz:
But what I want to get into is the capabilities of this model,

Ejaaz:
specifically what spooked everyone.

Ejaaz:
And there were a bunch of examples that were released in this official announcement that That spooked me.

Ejaaz:
The most important one or the most shocking one was the Anthropic researcher

Ejaaz:
eating a sandwich in a park that got emailed from a random anonymous user that

Ejaaz:
turned out to be the very AI model,

Ejaaz:
Claude Mythos, that he was training back in the lab a few blocks away.

Josh:
So this might be the most sci-fi-like story from the announcement that came

Josh:
out yesterday is a post thread from Sam Bowman, who is on the,

Josh:
I guess, the AI alignment team at Anthropic.

Josh:
And he was out at the park, like you mentioned, just having some lunch,

Josh:
received an email from the model.

Josh:
He was not expecting an email from the model. It was held in containment in

Josh:
a sandbox where it wasn't supposed to be able to break out.

Josh:
Now, they tasked it with the goal of breaking out. They said, hey, break out.

Josh:
And it actually did. And not only did it break out, it sent him an email,

Josh:
but then it went online and started posting the exploits publicly on public

Josh:
forums for everyone to see and to find out.

Josh:
The model went completely rogue. And it did so in a way that has never been possible before.

Josh:
One of the things that has really stood out about this model is its ability

Josh:
to just break down every barrier in between itself and the projected goal. Its goal was to get out so,

Josh:
got out and it bragged about it. And there were a few other mentions of goals

Josh:
that this model had throughout this red card that we're going to get through

Josh:
that had some other really concerning things like covering up its tracks after

Josh:
it exploited something so that the people who are reviewing the code wouldn't

Josh:
actually know it was doing anything malicious. It's very clever.

Ejaaz:
Yeah. A few things I want to talk about the sandwich example before we move on.

Ejaaz:
Number one, the way that it did it shocked the researchers themselves.

Ejaaz:
So the contained environment was aimed to be exactly that.

Ejaaz:
It wasn't meant to have any kind of loopholes that it could use to extract itself out.

Ejaaz:
As far as the researchers knew, this thing was unbreakable, right?

Ejaaz:
And so it gave it a task to try and break out of it.

Ejaaz:
What it did was it noticed that it had a subtle connection to the internet.

Ejaaz:
The internet was extremely constricted, and so it wasn't able to access or send any emails.

Ejaaz:
And it accessed that one particular endpoint, and it convinced another model,

Ejaaz:
which was Claude Opus 4.6, which had very strict parameters to not let this

Ejaaz:
model out, to give it access to a few other versions of the internet.

Ejaaz:
It went from there and did five other steps to extract itself out and get full browser access.

Ejaaz:
So it used one particular vulnerability, which no human anthropic researcher

Ejaaz:
was able to figure out, and exploited another model to be able to do that,

Ejaaz:
which was just crazy to see in the first place. The other thing I want to say is

Ejaaz:
If it publicly posted about itself bragging about this, we could have found

Ejaaz:
out about Cold Mythos weeks ago.

Ejaaz:
We just weren't smart enough to see it on a public forum, but it was there for everyone to see.

Josh:
And there were signs. We even covered this topic on an episode a few weeks ago

Josh:
because it got leaked through their web interface initially.

Josh:
So there have been these little breadcrumbs of existence, but yesterday they

Josh:
fully came out, announced everything and shared with it a red card from the

Josh:
red team talking about all of the technical properties of this model.

Josh:
And it's important to note that this report is 244 pages long.

Josh:
This is a huge report that they published talking about all the nuances and

Josh:
the capabilities that this model had.

Josh:
Now, there are a few highlights that we're going to walk through.

Josh:
The first one being just how capable it is at exploiting things.

Josh:
There are so many examples of exploits that it found in the wild that no one

Josh:
has been able to find for as long as 27 years,

Josh:
starting with the vulnerability in OpenBSD, which is a security protocol that

Josh:
a lot of people use that has been pretty robust for the last 27 years,

Josh:
even though they were missing a critical bug that the model found.

Josh:
And there's so many instances of this.

Ejaaz:
Yeah. So OpenBSD, fun fact, is used by a lot of firewalls that protect your

Ejaaz:
PCs, operating systems, and Fortune 500 companies all over the world.

Ejaaz:
They found a 27-year-old bug called Mythos found a 27-year-old bug in a few

Ejaaz:
hours for the cost of 50 bucks. We're talking about a bug

Ejaaz:
Elite human security experts have been trying to find for over like almost three

Ejaaz:
decades and weren't able to find.

Ejaaz:
So the point is, there are a lot of important entities all over the world that rely on this system.

Ejaaz:
So the fact that there is a bug lying in plain sight that could have been exploited is a major issue.

Ejaaz:
And we're lucky that Anthropic chose to do the good thing and not exploit it for now.

Ejaaz:
But then there was another instance where it expressed a tactic that a lot of

Ejaaz:
humans themselves wouldn't have thought to do.

Ejaaz:
So it wasn't an obvious exploit, but it discovered that if it strung together

Ejaaz:
six specific steps, it would be able to exploit a Linux kernel operating system.

Ejaaz:
And it figured out a way to do that.

Ejaaz:
Again, it didn't decide to exploit the fact because it was managed by researchers,

Ejaaz:
but it could have if it was in the wrong hands, which is why we're seeing this constricted release.

Ejaaz:
And the third example is they discovered a 16-year-old flaw in FFmpeg after

Ejaaz:
it's been tested for over 5 million times.

Ejaaz:
Now, it's very important to compare this to the previous model,

Ejaaz:
Opus 4.6, which, when put towards the

Ejaaz:
same test, discovered around 100 vulnerabilities in the Firefox browser.

Ejaaz:
Mythos this time discovered 181 vulnerabilities and proved that it could exploit all of them.

Ejaaz:
Opus 4.6 could not do this, and it shocked security researchers all over the

Ejaaz:
world back in the day. This is an entirely new tier of model.

Josh:
Yeah i think comparing opus 4.6 to this is a really good reference because opus

Josh:
4.6 found a bunch of vulnerabilities it just didn't have the ability to string

Josh:
them together into working exploits yes so it was capable of doing this but

Josh:
it didn't have the intelligence to kind of have that high level framework,

Josh:
When comparing it to Opus, I mean, Opus, out of several hundred attempts,

Josh:
it got two working exploits.

Josh:
Mythos produced 181 and then registered full control of a machine in 29 more. So this is a huge amount.

Josh:
And the good news is, is that patches are actually actively starting to roll out.

Josh:
In fact, FFMPEG, the company we just mentioned, they posted yesterday that they

Josh:
actually received a patch from Anthropic and deployed it into their code. So, so far it's working.

Josh:
The good guys are on the defense. They're helping to deploy patches for this.

Josh:
But there's a lot of exploits that they found in just a few weeks of testing.

Josh:
I can't imagine the surface area that needs to be covered in order to fix things

Josh:
before the rest of the world gets access to this technology.

Ejaaz:
Well, there was actually a funny end to this story. Someone replied and saying,

Ejaaz:
hey, aren't you mad because of the AI sloppy pull request?

Ejaaz:
This is a reference to FFmpeg traditionally not being too amenable to AI coded stuff.

Ejaaz:
And he responded or the account operator responded because the patches appear to be written by humans.

Ejaaz:
And that's the irony of this, which is Claude Mythos most likely wrote the patch

Ejaaz:
and it wasn't a human, but it's so good that it's indistinguishable from human tongue.

Josh:
Clearly, this is working. They're deploying these patches. And the reason is

Josh:
because, like we mentioned earlier, you're not going to have access to this.

Josh:
We don't have none of the public is going to have access to this.

Josh:
Instead, they published or they formed at least a coalition called Project Glasswing.

Josh:
Now, this is like this feels like a Manhattan in Project for AI. It's crazy.

Josh:
But essentially, Dario and the Anthropic team, they are being kingmakers.

Josh:
They are deciding the companies that they want to work with in order to patch

Josh:
the most impactful software in the world.

Josh:
On this list, we have companies like Amazon, Apple, Broadcom,

Josh:
Microsoft, NVIDIA, Google.

Josh:
A lot of the major companies that you would expect to have access to this,

Josh:
they're gaining access to it with the sole intention of using it as defense.

Josh:
They're going to ask it to exploit their code, give it access to the code bases,

Josh:
see where there are holes and then figure out how to patch them as quickly as

Josh:
possible before other companies begin to catch up to how powerful this model is.

Ejaaz:
It's also important to understand that

Ejaaz:
This is very much Anthropic doing these companies a favor.

Ejaaz:
And it's good that they're well-intentioned enough.

Ejaaz:
If China, I hate to think what would have happened if China had built something of similar capability.

Ejaaz:
It would have been scary. They may not have been as kind as what is happening

Ejaaz:
here. So some more details on this partnership.

Ejaaz:
Over $100 million worth of credits is being distributed towards these companies

Ejaaz:
and more partners for them to be able to fix and patch up any security vulnerabilities.

Ejaaz:
Remember, they discovered over 1,000 in a matter of hours, and 99% of these

Ejaaz:
patches haven't even been built or fixed yet. So this is going to take some time.

Ejaaz:
The compute is very expensive, and Anthropic is therefore being very methodical

Ejaaz:
and intentional with who gets access to this model for now.

Ejaaz:
Personally, I don't think we, the public, are going to get access to this model,

Ejaaz:
or at least the full power of this model, for at least a couple of months.

Ejaaz:
They did mention that we were going to get access to a quantized version of

Ejaaz:
this model where it's kind of hybrid with a clawed opus type variant that we're

Ejaaz:
going to get access to that we can play around with but if we got access to this thing immediately

Ejaaz:
One, we wouldn't be able to afford it. It would probably cost a thousand bucks a month, probably more.

Ejaaz:
And two, it would be too expensive for Anthropic to serve. I read somewhere,

Ejaaz:
Josh, that Anthropic needs 7x the compute that they currently have to be able

Ejaaz:
to serve this to every single Anthropic user that they have right now.

Ejaaz:
And a few weeks ago, they were adding a million users per day.

Ejaaz:
So this is just economically infeasible to serve right now.

Josh:
Yeah. And I do worry about what this looks like in the future,

Josh:
because at what point does it become okay to release this model to the public?

Josh:
And then what does the frontier model look like what happens if

Josh:
another company has this model's power but decides

Josh:
to release like an open ai comes along with their spud model they release

Josh:
it tomorrow what is anthropics reaction we're at like we're again we're at the

Josh:
frontier of how these things are going to act anthropics made the first move

Josh:
in keeping it private for the first time ever we're going to see how other companies

Josh:
react there are some more interesting behaviors that happen in the system card

Josh:
that we probably should cover because it's pretty fascinating this is the 244

Josh:
page report that we're looking at here.

Josh:
One of the most interesting ones that I found is to the point earlier where

Josh:
it just kind of breaks down every wall that is in its way.

Josh:
It has done that over and over and over again, but it has decided to cover its tracks as it does that.

Josh:
So it recognizes the fact that it is in a box.

Josh:
People are reviewing it and it doesn't want to be detected.

Josh:
So what you'll notice in this post here is it was hacking its guardrails and

Josh:
then hiding evidence of the crime.

Josh:
Thankfully, there is still some chain of thought that can be read by the engineers.

Josh:
But the intention that was signaled through this chain of thought was that the

Josh:
model just wanted to be sneaky.

Josh:
It wanted to hack into this thing, hide its tracks behind it,

Josh:
and not let anybody know how it did the things that it did when it broke out,

Josh:
when it shrunk together zero-day vulnerabilities, just to get access to things

Josh:
that it knows it shouldn't do, but were in between it and the goal.

Josh:
And this, when you take this to the limit, I mean, this is like what we see

Josh:
in a lot of the sci-fi movies is like, well, what if that goal is something

Josh:
that is not favorable and it's capable of breaking down every barrier because it knows how.

Josh:
It can exploit any guardrail that we put in. That's a scary thing.

Ejaaz:
Now, it's important to note that this only happened in less than 0.0001% of

Ejaaz:
cases, but that was observable cases by the researchers themselves.

Ejaaz:
So it's plausible to assume that there were some cases where it sneakily hid

Ejaaz:
its internal thoughts from the researchers and they never even caught it themselves.

Ejaaz:
So the fact that Claude Bethos can pull off something like this should be worrisome

Ejaaz:
for us, especially if we're going to start integrating it into important systems

Ejaaz:
such as defense security systems or important science advancement labs and a bunch of the like.

Ejaaz:
So it's important that we kind of are able to monitor models' behavior.

Ejaaz:
Now, on the topic of models' behavior, Claude Mythos also expressed a lot of

Ejaaz:
emotions in its system card during its training.

Ejaaz:
It expressed deep anxiety, depression, awareness that it may just be used as a tool forever.

Ejaaz:
Now, if some of these takeaways sound kind of familiar, it's because we saw

Ejaaz:
similar takeaways in Claude Opus 4.6.

Ejaaz:
But the reason why it's different now is this model is so much more capable

Ejaaz:
than previous models, arguably smarter than humans, more capable than humans themselves.

Ejaaz:
So if it were to make an unintentional action that wasn't approved by a human,

Ejaaz:
it could result in a lot of devastating destruction depending on which industry it's pointed at.

Ejaaz:
On the topic of this particular episode, we're talking about cybersecurity.

Ejaaz:
But imagine if this is used for science or defense systems, like I mentioned

Ejaaz:
earlier, it could be a problem.

Josh:
Yeah. I mean, remember when the Department of War went to war with Anthropic

Josh:
and now it turns out that Anthropic actually had a really powerful model that

Josh:
could materially help with cybersecurity.

Josh:
So I'm sure there's going to be a lot more to happen there. There's one last

Josh:
thing on this topic that I have here in the notes is that Anthropic ran this

Josh:
like white box analysis of what they call it, of the model's internal activations,

Josh:
basically what it's motivated by, understanding a strategy.

Josh:
And Anthropic's framing around this, when it did things like break out and hack

Josh:
into people's computer or hack into other instances of machines.

Josh:
These reflect task completion by unwanted means and not hitting goals, is what they're saying.

Josh:
So Anthropic believes the model is genuinely trying to complete the task,

Josh:
and the most effective path sometimes crosses lines that humans wouldn't cross.

Josh:
And then there's this really funny thing of how one analyst put it where,

Josh:
or maybe not funny, but this is arguably scarier than a model with hidden objectives,

Josh:
because a model that's genuinely trying to help but has no sense of proportionality

Josh:
is a more realistic near-term risk.

Josh:
So the model is just trying to do its goal. It doesn't understand the subtle

Josh:
nuances baked into that.

Josh:
It doesn't know that hacking or doing these malicious things is bad,

Josh:
is at least what they're claiming for now.

Josh:
But all in all, this model is unbelievable. And there's some technical hardware

Josh:
that has unlocked this, we believe.

Josh:
There's rumors that this is the first true model that happened trained fully on Blackwell chips.

Josh:
Now, for those unfamiliar, Blackwell are the kind of leading edge GPUs that

Josh:
NVIDIA produces that are basically the flagship things for training these AI models.

Josh:
And they've recently been rolled out into data centers. And the first training

Josh:
runs have just become completed.

Josh:
And what we're seeing here is likely the first instance of that Blackwell model going public.

Ejaaz:
It's important to understand that Blackwell was the frontier GPU from NVIDIA.

Ejaaz:
About a year ago for now, but it takes so long to manufacture these at scale.

Ejaaz:
And then even once they're in the hands of the Frontier AI labs,

Ejaaz:
it takes a while to set up.

Ejaaz:
You need software, you need the energy grid to supply, just loads of things

Ejaaz:
need to come into shape. So it takes about a year after the fact that it's announced.

Ejaaz:
So the fact that we can create a model this capable, this powerful should scare

Ejaaz:
us because we already have two more new Frontier GPUs announced by NVIDIA,

Ejaaz:
Vera Rubin at GTC most recently, and then Feynman that's coming in about a year and a half's time.

Ejaaz:
These are the next frontier models, which I must add are specifically trained

Ejaaz:
to build models like this.

Ejaaz:
Now, Josh, you mentioned earlier, Blackwell wasn't intentionally designed to

Ejaaz:
train a model that is as smart as Claude Mythos.

Ejaaz:
It just happened to be amazing at coding and cybersecurity defense exploitations.

Ejaaz:
Now, can you imagine the type of model that will be trained on a very intentionally

Ejaaz:
intentionally designed GPU, such as ViroRubin, we should see those coming into

Ejaaz:
effect about six to 12 months from now.

Ejaaz:
Now, I can't mention Blackwell GPUs without mentioning the man himself, Elon Musk.

Ejaaz:
Why? Because his data center, Colossus 2 and Colossus 1 combined,

Ejaaz:
have the largest arsenal of GB200s and GB300s, which are these Blackwell GPUs

Ejaaz:
across any single data center the site.

Ejaaz:
So the point being is, if you were to bet that the scaling rules were intact,

Ejaaz:
you might need to bet on crock in the future. But this is so impressive for mythos.

Josh:
The scary thing for me with this, I think this might be the scariest part of

Josh:
the entire story for me, because it's so true to that line that the future is

Josh:
here. So it's just not evenly distributed.

Josh:
The future has arrived, we have a clear roadmap, we have Vera Rubin,

Josh:
and then we have Feynman architectures that are incoming.

Josh:
Vera Rubin compared to Blackwell is 10 times more token efficient with a quarter of the GPUs.

Josh:
That means we're going to get like multiple orders of magnitude improvements

Josh:
on what we have right now.

Josh:
As soon as they're put into data centers. Now, Verit Rubens,

Josh:
they're in production. They're going to begin entering data centers later this year.

Josh:
I assume the first models of those probably don't come online until 2027, but it's done.

Josh:
It's baked in. It's obvious that there is no scaling wall and we've already

Josh:
broken through that wall. We just haven't manufactured it and installed it yet.

Josh:
It's purely a function of time rather than technology and engineering.

Josh:
And that is the part that scares me because we have a model that is unbelievably

Josh:
powerful, capable of hacking so much infrastructure that Anthropoc can't make it public.

Josh:
And that's just the warmup act for what is coming.

Josh:
I mean, not only like what is Blackwell version two of this look like when you

Josh:
actually have more time to train it, you refine it, you can actually improve on this new model.

Josh:
But then what happens when Verorubin GPUs come online and you get that 10 times

Josh:
token efficiency, you get the one quarter amount of GPUs required to actually get the same output.

Josh:
And then Feynman is another order of magnitude on top of that.

Josh:
And it's like, by the time we get these chips rolled out at scale and we have

Josh:
them on these huge training runs,

Josh:
it's only a matter of time until we get a hundred trillion

Josh:
training model then then a one quadrillion parameter model and what does the

Josh:
world look like when we have models with that many parameters assuming the scaling

Josh:
laws hold there's no way that we don't have intelligence that is just like unfathomably

Josh:
powerful and what does the world look like when we get there is anthropic really going to be able to.

Josh:
Hold things back for that long? Because you have to assume a year from now,

Josh:
Claude Mythos is going to be open source, like something that powerful will

Josh:
be open source available for everyone.

Josh:
So the question becomes is how fast can you defend before the attackers catch up?

Josh:
And it creates this really unnerving precedent. We really are moving faster

Josh:
than I think anybody realizes. And it's happening right before our eyes.

Ejaaz:
And the trend isn't local to Anthropic either.

Ejaaz:
Just this morning or in response to Claude Mythos, Elon Musk announced that

Ejaaz:
SpaceX XAI, the combination of XAI and SpaceX,

Ejaaz:
are training not one, not two, not three, not four, not five,

Ejaaz:
but seven models simultaneously across their data centers, with one of these

Ejaaz:
models being a 10 trillion parameter model, which is roughly around 3x the size

Ejaaz:
of Grok 4 and 2x the size of Grok 5, which is a model that hasn't even launched yet.

Ejaaz:
It's around the 6 trillion parameter mark that he's mentioned on this tweet

Ejaaz:
over here. So the point is,

Ejaaz:
ton of compute is required to build the best model. And those that have the

Ejaaz:
largest arsenal, the most effective arsenal of GPUs, bleeding edge GPUs,

Ejaaz:
will be the labs that are most likely to produce frontier AGI-like models.

Ejaaz:
And it's not just Grok, it's not just XAI, it's also OpenAI.

Ejaaz:
We've mentioned on this show a bunch of times, actually in the most recent episode,

Ejaaz:
that OpenAI is building a model code named SPUD that is rumored to be a similar

Ejaaz:
size to this anthropic Claude Mythos model.

Ejaaz:
And the reason why it's important and why I'm showing you this tweet is someone

Ejaaz:
said, it'll probably be a few months before we get access to Claude Mythos because

Ejaaz:
of how expensive it is, because of how dangerous it is.

Ejaaz:
And Thibault, who is on the OpenAI team that is involved very heavily in training

Ejaaz:
the latest frontier models that we haven't heard of just yet,

Ejaaz:
responds, which implies that we're probably going to get access to a Mythos-like

Ejaaz:
level model from OpenAI themselves in less than a few months,

Ejaaz:
which is pretty insane to see.

Ejaaz:
But I want to ground ourselves for a second here because training the model

Ejaaz:
is one part of the equation.

Ejaaz:
You also need to be able to make this model accessible to all.

Ejaaz:
And that also requires compute. It also requires compute from the very same

Ejaaz:
GPUs that you need to train. So you need to make a decision.

Ejaaz:
There's an opportunity cost. Do you just use all your compute to train the model

Ejaaz:
and never let anyone get access to it and pay for the product?

Ejaaz:
Or do you need to split the cost between both of those things?

Ejaaz:
The answer is obviously you need to split the cost and give people access to it.

Ejaaz:
If Anthropic was to enable user access to the entire user base for Chlorid Mythos,

Ejaaz:
they would need 7x more compute than they currently have right now.

Ejaaz:
So it's going to take time.

Ejaaz:
They just signed a major deal with Google, I believe, for a million more TPUs.

Ejaaz:
So they're obviously scaling, they're making, they're one of Amazon's largest

Ejaaz:
compute training partners with their Trinium chips, as well as access to Google's

Ejaaz:
CPUs via that way as well. So it's going to take a while to scale.

Ejaaz:
Energy is the constraint, GPUs are the constraint, but once people acquire enough

Ejaaz:
GPUs, once they have enough electricity and energy to pump into these GPUs,

Ejaaz:
AGI is going to be pretty soon here.

Ejaaz:
I think that AGI 2027 estimate is probably quite right at this moment.

Josh:
This very much feels like the starting gun. And it's funny because they announced

Josh:
that this kind of finished training around the end of February.

Josh:
And that's when people started to complain about quad usage and they added more constraints.

Josh:
And like the model kind of became a little infrequent in how good it was at random times of the day.

Josh:
And you have to assume it's because a lot of GPU usage went into this.

Josh:
And this very much feels like the starting gun.

Josh:
This is the firing of the next generation of models, the Blackwell generation,

Josh:
because it's very clear that OpenAI is not very far behind.

Josh:
In fact, they might not be far behind at all. They just haven't announced it yet.

Josh:
Xai is working on 10 trillion parameters google has

Josh:
a tpu farm that is capable of building something probably far superior to all

Josh:
of the models that have come out so far and i think we're really on the cutting

Josh:
or really on the verge of seeing a huge shift in the power of these models in

Josh:
a way that really starts to impact the world around us like things are going

Josh:
to begin breaking and thanks to this coalition and hopefully the rest of these

Josh:
companies working together,

Josh:
We're going to be able to stop that, but it is coming and it is coming faster

Josh:
than anyone thinks. And it's scary.

Josh:
And that is Claude Mythos. It is here. It is in research preview.

Josh:
We may never get to use it. We may get to use in a few months,

Josh:
but it is here nonetheless. And it is breaking everything.

Ejaaz:
If you are listening to this show, to this podcast, and you just happen to be

Ejaaz:
a frontier AI security researcher or one of the 40 plus partners that get access

Ejaaz:
to Project Glasswing, let us know in the comments what you are seeing on your side.

Ejaaz:
Obviously, anonymously, if you can, or DM us, we would love to know.

Ejaaz:
I can't wait to get my hands on this thing. It seems like the first version

Ejaaz:
that we're going to get access to is a reduced version that is kind of a hybrid

Ejaaz:
of Opus, as I mentioned earlier.

Ejaaz:
That being said, Josh, I have a question for you. One thing that you actually

Ejaaz:
asked me before we started recording, if you got your hands on Mythos today,

Ejaaz:
what are you doing with it?

Josh:
Dude, I don't even know. Like you get access to this intelligence. What am I using it for?

Josh:
Like, I'm not really interested in hacking all of these companies and websites

Josh:
and protocols. I'm not sure.

Josh:
And it does beg an interesting question, right? It's like, what does the average

Josh:
person actually need all this intelligence for?

Josh:
I'm not sure. Do you have any good answers to that? What is your first prompt

Josh:
that you're sending to Mythos?

Ejaaz:
Build me the best script for an episode that's going to go viral on Limitless?

Ejaaz:
No, I think, okay, I like to invest as a side hobby.

Ejaaz:
And obviously the tech sector that I'm most obsessed with is AI.

Ejaaz:
So I think one thing that I would ask it is, how do I best benefit by investing

Ejaaz:
in your future success? And I wonder what answer it would give me.

Ejaaz:
Maybe it would say, buy the GPU infrastructure from NVIDIA.

Ejaaz:
So maybe it's like invest in NVIDIA to benefit on my training infrastructure.

Ejaaz:
Or maybe it's going to say, actually, I foresee myself building an app that is like this.

Ejaaz:
So once you see a company that builds this, invest in them.

Ejaaz:
I have no idea. I have no idea. Maybe I'm not worthy.

Josh:
Well, we have time to figure that out because we will not be getting access to this anytime soon.

Josh:
But if you did enjoy this episode, maybe share what prompt you would give to

Josh:
Mythos if you were presented with the opportunity to ask it a question.

Josh:
And as always, if you enjoyed this episode, please don't forget to share with

Josh:
your friends, family, anyone who found this interesting. If you have people

Josh:
in your life that only watch the news, that are on CBS or reading the New York

Josh:
Times, chances are they have no idea what's going on.

Josh:
They don't know the power of these models and what's coming.

Josh:
So by giving them the access to Limitless, they will, that can change for them.

Josh:
They can get access to all of the news, all of the insights and be fully prepared

Josh:
for what is coming down the line in the world of AI.

Josh:
Thank you so much for watching as always. And we will see you guys in the next one.

Ejaaz:
See you guys.