The AI Model Too Dangerous to Release

Dear Reader,

This is Dylan Jovine with Behind the Markets.

Happy Tuesday.

Today is Tuesday, April 14th.

There is so much to talk about today…

Let me start with this startling announcement from Anthropic.

The AI Model Too Dangerous to Release

Last week, Anthropic released a note saying their newest AI model is too dangerous to release to the public.

During testing, they found that the model — code named “Mythos” — is capable of identifying and exploiting zero-day vulnerabilities in every major operating system and every major web browser.

In plain English, Anthropic's newest AI has found security holes in the systems that power basically everything.

Computers, banks, the stock exchange, Apple, Microsoft, Bank of America.

It's crazy.

What That Actually Means

They found thousands of security vulnerabilities that were previously unknown to software developers.

Some of these vulnerabilities date back more than two decades.

Think about how that happens.

Microsoft builds core code.

Then someone builds on top of it.

Then someone builds on top of that.

Banks have code written many decades ago with layers and layers of new code stacked on top of it.

You tell this AI to go test Chrome or go test a bank's website and find vulnerabilities.

And it does.

In eight minutes, according to one account.

Was This Just Hype?

I'll be honest — my first reaction was skepticism.

You want to beat ChatGPT in the AI race.

You want attention.

Saying "our model is too powerful to release" is a pretty good way to get it.

But then something changed my mind.

Treasury Secretary Scott Bessent and Fed Chair Jerome Powell took this seriously enough to pull in the heads of Apple, Google, and the major banks for an emergency meeting.

That's not a PR stunt.

That's a real warning being taken seriously at the highest levels.

What Anthropic Is Doing About It

Anthropic has now launched something called “Project Glasswing.”

The idea is to work with tech firms and critical infrastructure — banks, software companies, water systems, power companies, healthcare — and run this AI through their systems before it ever gets released to the public.

I am so glad they are doing this.

It gives our side a chance to test and harden our critical infrastructure before bad actors get their hands on something like this.

And frankly, it gives our government a powerful tool to use against our national security competitors — whether that's Iran, China, Russia, or whoever.

The Bigger Question This Raises

If one AI model can find major security vulnerabilities in eight minutes, what does that mean for cybersecurity firms?

Do you even need them anymore?

Or can AI do the job faster, cheaper, and better?

This brings us back to what a lot of people are calling the SaaS apocalypse.

The idea that AI has the potential to eat the lunch of entire categories of software companies.

What We're Seeing Here at Behind the Markets

I can tell you firsthand — we are using AI to do things we used to pay vendors for.

Good vendors, filling real needs.

But we're finding that AI lets us shave significant costs off our income statement and redirect that money toward building better products for you.

You've already seen some of this with the tools we've released to Alliance members.

We have one out now that helps you generate income from your portfolio.

We have another coming that approaches income from a different angle.

And we have another one after that.

The threat to traditional software firms is very, very real.

But for us — and for you — the lower costs are creating real opportunities.

And frankly, this is exactly the kind of moment I love as an investor.

When a new technology disrupts an entire category, the companies supplying the picks and shovels tend to win no matter who survives the chaos.

Anyway, that is all I have for you today.

Have a wonderful Tuesday.

I will see you tomorrow.

“The Buck Stops Here,”