An AI model too powerful for public use? That's how Anthropic is selling its latest advanced AI model ethos.

Kate Rooney has the story for today's tech check. Kate.

Hi, Sarah. So this really is unprecedented. Anthropic is going to be restricting public access to start.

Its newest AI model is only available to a handful of tech partners who the company says can get what they

describe as a head start on any bad actors out there with nefarious plans for ramped up versions

of this technology.

It's going to be called Claude Mythos previews.

So Anthropic CEO Darya Amadeh says that they actually trained it to be good at code, but as a side

effect, it's also good at cyber.

It was, it is rather rolling out to 40 companies.

You've got Apple, Google, Microsoft, Nvidia, AWS called Project Glass Wing.

Anthropic says the model already has autonomously found thousands of high severity vulnerabilities,

including in every major operating system and web

browser. They even cite some examples that were 30 years old guys. So they also warn of

certain vulnerabilities in corporate networks out there, health care systems, energy infrastructure,

and then state-sponsored attacks from Iran and other adversaries. Cybersecurity stocks did take a hit on that initial

report that we saw that this model was going to be coming out. But today, it's actually the

opposite. Some of its partners are getting a boost. You've got Palo Alto networks and crowd strike

that are part of this initial rollout. J.P. Morgan calling out those partners as beneficiaries

calls this model a significant leap forward versus prior model says that these cyber names are, quote,

essential layers in the defense stack rather than competitive targets in this effort.

They describe it as AI or fighting AI with AI.

...