What Claude Just Did Is Insane (Investors Aren t Ready)

Money Moves
6 followers
1 Video View·Apr 16, 2026
Comments are disabled for this video.
An AI model just found serious security flaws in every major operating system and web browser
on earth. One flaw was hiding for 27 years inside one of the most secure systems ever built.
Automated tools scanned that code 5 million times, and they never caught it. But here's
where things get really crazy. The company that built this AI decided that it was too dangerous
to release. So instead, they gave it to 12 of the most powerful companies on the planet,
almost all of which we can invest in. My name is Alex, and I spent 8 years as a radar engineer
and AI researcher at MIT, and I've never seen AI do anything like this. So subscribe to the channel,
and let me show you what just happened, and how I'm investing in it. Your time is valuable,
so let's get right into it. There's a piece of software that's so foundational that it powers
almost every video you've ever watched online, including this one. It's called FFmpeg,
and it had a bug. It used two different kinds of counters that didn't quite line up under one
very specific condition. A frame carefully split into exactly 65,536 tiny pieces. That's 2 to the
16th power. Those counters would collide when that happens. And when that happened, the software
could write data outside of the memory it was allowed to touch, which opens the door for
to take control of the machine running the video. That's your PC or your smartphone if you're the
one playing that video. But it's also the enterprise servers that are processing it.
Netflix. Apple. YouTube. That bug was introduced in 2010. For 16 years, automated testing tools
saw that same line of code. 5 million scans. Every single one missed it. And the reason is simple.
Traditional tools throw random inputs at software and wait for it to crash. But this bug only
triggers at one specific number. A random number tester would almost never find it. But this AI
read the code, found the flaw, generated a custom test for it, and confirmed the bug on its very
first try. This isn't a specialized cybersecurity tool. It's not a model that's trained to hunt
vulnerabilities. It's a general-purpose language model called Claude Mythos, built by a company
called Anthropic. And here's why that matters for investors. Anthropic didn't set out to build a
hacking tool. They set out to build a better coding assistant. But the model got so good at reading
code and so good at understanding what a programmer intended, as well as spotting the difference
between the two, that it could find the flaws that human experts have been missing for decades.
So Anthropic trained Claude Mythos to be really good at code. But as a side effect,
it got really good at securing code as well. To understand why this is such a huge deal for
software stocks, you need to understand how Mythos actually finds these issues.
The model is placed in an isolated environment with access to a specific codebase. It reads
through the source files, which is millions of lines of code, and finds the parts most likely
to be hiding serious bugs. Then it writes and runs test programs against that code to confirm
or disprove each potential bug. When it finds one, it writes a formal vulnerability report
with a small example that recreates the flaw in practice. That's exactly what a professional
security researcher does today. The difference is speed and scale. A human analyst might take
weeks to audit a single block of code, especially if it's mission critical. Mythos processes an
entire codebase in hours. And it doesn't just find bugs, it understands them. It reads the code
the way an engineer reads a blueprint, understanding intent, spotting where the
code doesn't match that intent, and reasoning about what would trigger the failure. What's
even crazier is just how fast this happened. Cybergym is a UC Berkeley benchmark that
throws over 1,500 real-world software bugs from almost 200 open-source projects at different AI
models. The current Claude model, Opus 4.6, scored a 66%. Claude Mythos scored an 83. That's a 16-point
leap in one model generation. The difference between a D and a B on a test. That's the jump
from occasionally finding bugs to finding bugs in every major operating system and web browser
on Earth. That's what makes this model so dangerous. But it didn't stop there. It found a 27-year-old
bug in OpenBSD, which is an operating system that prides itself on being almost impossible to hack.
This one matters to investors for two reasons. First, we talk a lot about AI networking,
but not the code that makes it possible. And second, this wasn't a single bug, it was two
bugs that chained together. The first was a missing safety check, and the second was a number rolling
back to zero at the wrong time. When combined, they could purposely crash a machine from anywhere
on the internet with no authentication required. Governments have trusted this system to run their
firewalls since 1998. Mythos found this bug in 2026. It also found flaws in FreeBSD, which is
how Firefox runs JavaScript. It found over 180 different bugs in everything from cryptography
libraries to virtual machine monitors, and turned them all into functional attacks. Previous models
found just two of these bugs. And on Linux, which runs most of the world's servers,
Android phones, and cloud infrastructure, it chained four separate vulnerabilities together.
On their own, each bug looked pretty insignificant. But combined,
they let an attacker jump from an ordinary user account to full control of the Linux machine.
Mythos is not a proof of concept. Mythos is a weapon. And that's why it scared the market.
But there's something else on the market that you need to know about, and that's your private data.
There are hundreds of online data brokers making big money by collecting and selling
your personal information. That's why I've been using this video's sponsor,
Delete.me, for over two years now, and I really can't recommend them enough.
Delete.me is a hands-free subscription service that will remove your personal information from
those online data brokers. They give you a quarterly privacy report showing everything
they've done. And they reviewed over 55,000 listings for me so far. But what really surprised
me is these data brokers had way more than just my private data. They had my wife's,
and my entire family's too. That's another reason I really like Delete.me. They have a family plan
so we can all have more control over our personal data. So if you care about your data and your
family's privacy, you can get 20% off any consumer plan with my code, symbol20, by going to join
delete.me.com or with my link in the description. And a big thank you to Delete.me and to you for
supporting the channel. Alright, the AI arms race is moving fast. In 2024, Google's project
Big Sleep found a single new vulnerability in SQLite. In 2025, a DARPA competition threw 54
million lines of code at competing systems that collectively found 18 bugs. In April of 2026,
Mythos found thousands across every major operating system, every major browser, and every
major code library on the planet. And it didn't just find bugs, it built working attacks around
them. Anthropic says that over 99% of these bugs are still unpatched, because the fixes just haven't
been deployed yet. And that should have the market's full attention. And it did, but probably
not the way Anthropic wanted. Two weeks before they were ready to announce Mythos, a blog
misconfiguration leaked it to the public. And the market was very quick to react. Cybersecurity
stocks tanked. CrowdStrike. Palo Alto Networks. Everyone. The logic was obvious. If an AI model
can find bugs, exploits, and vulnerabilities faster than any human, then why do we need a
quarter trillion dollar cybersecurity industry in the first place? The market didn't see Mythos as
a competitive threat. It saw it as a meteor. And expensive cybersecurity consultants were the
dinosaurs. And for a few days, it looked like the market was right. The question was which company
would use these capabilities first. And that's where the next phase of this AI arms race begins.
Alex Stamos is the former chief security officer at Facebook and Yahoo. Today, he's the chief
product officer at Corridor, an AI security startup. Alex put a number to the timeline.
Six months. That's how long before small, open source models can find bugs as well as Mythos.
And once that happens, every cybercrime syndicate, every state-sponsored spy,
and every individual hacker on the planet gets access to AI-powered exploit discovery.
And here's the uncomfortable truth for this entire industry. The bottleneck was never finding the
exploits. It was never about bug discovery. Arctic Wolf's threat report found that 76%
of actual compromises – the real breaches, the real data being stolen, the real ransomware being
deployed – involved one or more of just 10 known already-patchable vulnerabilities. The flaws were
already found. The patches already existed. Organizations just could not move fast enough
to deploy them. AI just removed the speed limit. Not for cybersecurity companies, but for the
attackers. When every bad actor on the planet can find bugs at AI speeds, the gap between
this bug exists and this bug was exploited collapses from months to hours. The entire
defensive model of the cybersecurity industry – find it, report it, patch it, and deploy it – assumes
humans are the bottleneck on both sides of the equation. That assumption just broke, since
attackers can sit at home and point AI models at public software pretty much as fast as they can
type, while defenders have to fix the code, test it, get approvals, schedule maintenance windows,
and avoid breaking their customer deployments – that means tickets, meetings, and manual reviews.
On top of that, attackers benefit from automation – scanning for weaknesses, writing phishing emails,
exploring attack paths, and generating exploit code, while defenders have to slow down to make
sure they meet regulatory requirements, work with legacy systems, check with multiple vendors,
and have compliances that they need to uphold. You can't just let an AI patch a problem at a bank
or a hospital and hope for the best, but you definitely can attack them that way.
And that 6-month window is actually shrinking. ILE, an independent security research lab,
tested 8 smaller, cheaper, publicly available models to see if they could reproduce Mythos'
findings. All 8 models found the same exploits that Mythos did. One of those models had 3.6
billion parameters and cost $0.11 per million tokens. For investors, that means it's not
about the model itself. It's about the systems built around it – the targeting, the validation,
the triage, the relationships with maintainers – but the raw capability is spreading fast.
In fact, Palo Alto Networks' chief security intelligence officer said open models are only
weeks or months away from matching Mythos. Cybersecurity trainers at the SANS Institute
say that that capability exists now, for finding and exploiting basic vulnerabilities.
So 6 months is the optimistic estimate, and the clock started on April 7th.
Now let me say the quiet part out loud. The AI model finding bugs to patch is the same AI model
that's exploiting them. There's no architectural difference. There's no switch you flip from
offense to defense. It's the exact same capability that makes Mythos a shield that also makes it a
sword. Anthropic's own red team said it best. The same improvements that make the model
substantially more effective at patching vulnerabilities also make it substantially
more effective at exploiting them. Before going public with Mythos,
Anthropic briefed senior officials at two agencies responsible for defending America's
digital infrastructure. NSA analysts were already discussing what Mythos means for
cyber operations. And as an investor, think about what this means for the market. Anthropic,
a private company valued at $380 billion dollars after the second largest venture capital raise in
history, is sitting on software exploits for almost every major company on earth.
Over 99% of the vulnerabilities that Mythos found are still unpatched, which means Anthropic
is effectively deciding what gets fixed first, how much information to release, when and to who.
This is not a product launch. It's a national security threat. And the real question is who
ultimately controls this capability, who gets access besides Anthropic, and what happens when
open models start finding these same exploits all over the internet. Those aren't questions you're
going to get answers to on an earnings call. And this is where Anthropic made a big decision that
affects the entire stock market. They didn't sell Mythos to the government. They didn't
license it to the highest bidder. And they didn't open source it to level the playing field.
They formed a defensive coalition called Project Glasswing and gave early access to some of the
most powerful publicly traded companies on the planet. Amazon, Apple, Broadcom, Cisco, CrowdStrike,
Google, JPMorgan Chase, Microsoft, Nvidia, Palo Alto Networks, and more than 40 other organizations.
The cybersecurity stocks that went down on the leak surged on the announcement. CrowdStrike
jumped over 6% in a single session. Palo Alto rose almost 5%. The Mythos model went from being
a sword pointed at the industry to the shield protecting it. And now that you understand what
just happened, here's exactly how I'm investing in it. The biggest thing for investors to understand
is that the cybersecurity industry is about to undergo a massive shift, from detecting and
responding to threats to predicting and preventing them. The cybersecurity industry spent the last
two decades building tools to catch attackers after they got in. Glasswing is the first serious
attempt to find every flaw before the attackers do. And the cybersecurity companies inside this
coalition are not being disrupted by Mythos. They're being armed with it. If it works,
this will be the single biggest shift in cybersecurity since the invention of the
firewall. And if it doesn't, every enterprise on earth just entered an arms race, they have no way
to win. But let's talk about who wins from this today. The obvious winners are the companies inside
Project Glasswing, the ones with direct access to Mythos, and the infrastructure to deploy what it
finds. CrowdStrike focuses on endpoint detection and response, so securing desktops, laptops,
and smartphones, really any device that connects to a company network. Their Falcon platform has
three parts, a suite of cloud-based modules that do things like run virus scans, manage firewalls,
and detect malware. They have a separate threat graph that maps out a company's networks,
the devices on it, the users, and the permissions to make sure all the network traffic is legit.
And then their Falcon agent is a lightweight agent that runs on each device to send security
data back for analysis. Charlotte AI is their AI assistant layer, and Agentworks is their
agentic automation platform. Being armed with Mythos means that CrowdStrike can use their
platform to proactively patch vulnerabilities across their entire customer base before
attackers find them. CrowdStrike's revenue came in at $1.31 billion for their latest quarter,
which is up 23% year over year. Annual recurring revenue came in at $5.25 billion,
which is up 24%. Net new ARR came in at $331 million, up 47%. And if you think you missed
the boat on CrowdStrike, they just posted their first-ever GAAP profitable quarter,
and their gross customer retention sits at 97%. CrowdStrike's stock is not cheap. A $100 billion
market cap means roughly 20 times price to sales. But now that they're armed with Mythos,
we can see their revenues, their profits, and their overall growth continue to accelerate,
while their competition still waits for access. Palo Alto Networks is the other pure-place
cybersecurity company in Project Glasswing. Their strategy is convincing enterprises to
consolidate their security spending onto a single platform, network security, cloud security,
access protection, and so on. Cortex, their AI-powered security operations platform,
integrates directly with Mythos for proactive threat detection and response. Palo Alto's
annual recurring revenue from next-generation security grew by 33% year over year, and their
guidance implies more than 50% growth in next-gen security for the rest of this fiscal year.
Their total annual revenue is approaching $11 billion, which makes Palo Alto Networks
the cheaper stock. They trade at roughly 14 times sales compared to CrowdStrike's 20,
mostly because CrowdStrike is growing significantly faster. The hyperscalers,
Microsoft, Google, and Amazon, form the platform layer of Anthropx Project Glasswing. They host
Mythos, they use it internally, and they'll probably add it to their server-side security
offerings over time. Microsoft alone runs the largest cloud security business on the planet,
bringing in roughly $28.5 billion a year. These companies aren't using Mythos to sell more cloud
access, they're using it to manage risk, across the entire infrastructure powering AI and the
internet. The best way to find great investments is understanding a company's products, not just
their profits, and the best companies have perfect products for quickly growing markets.
Nvidia, Broadcom, Apple, these companies armed with Mythos will define the next era of digital
security, while everyone else is effectively defending against tomorrow's attacks with
yesterday's tools. And AI is already making the global cybersecurity market grow fast,
from $380 billion in 2026 to $1.2 trillion in 2034. That's a 15.5% compound annual growth rate
for the next 8 years, faster than the growth of the S&P 500. And I expect it to accelerate
as more crime syndicates, more spies, and more hackers in general start using more advanced AI
for their attacks. So, the question for investors isn't whether cybersecurity spending will keep
climbing, it's which companies will capture it. That's why the ones in Project Glasswing
are the ones I'm investing in. But here's the part we can't model in our spreadsheets.
Less than 1% of the thousands of vulnerabilities that Mythos has discovered
have actually been patched. Less than 1%. Anthropic promised a public report in the
next 90 days. That's July 2026. And in it, they're going to spell out what they've found,
what's been fixed, and who is still exposed. If that report shows 10-20% of the bugs have
been fixed, then the defenders really do have an edge. But if it only shows 1% of the bugs have
been fixed and the vulnerabilities have been closed, then the critics will be right. Attackers
will move as fast as AI, while defenders will stay at the speed of tickets, the speed of meetings,
and manual reviews. Remember, finding bugs can be automated, but fixing them can't,
especially for banks, for hospitals, and for other regulated industries.
That's why I think Palantir will play an important role here too.
But there's one last conflict that I'm not sure how Anthropic will overcome, or if they even can.
Anthropic is reportedly considering an IPO in October of 2026. That means that the company
that decided Mythos was too dangerous to sell will also need to justify a half trillion dollar
valuation to public market investors. The conflict between cyber safety and shareholders
is very real. Anthropic's 90-day security report will land in July. Open models are improving every
single month. The clock is ticking. But whether Anthropic's bet that six months of Mythos running
defense can outpace AI-powered offense will pay off for the companies in Project Glasswing,
for their stocks, and for the security of every data center on the planet,
that's something the market hasn't priced in yet. Let me know what you think in the comments.
Is Mythos a temporary edge for the defenders, or is this AI-driven arms race the start of a
new era of cyber security? And if you want to see more science behind the stocks,
check out this video next. Either way, thanks for watching and until next time,
this is Ticker Symbol You. My name is Alex, reminding you that the best investment you can
make... is in you.
