This is The Guardian.

Earlier this month, Anthropic, the company behind the clawed family of tools,

announced a new AI model.

And as with those that came before it, the buzz and hype began immediately.

But Anthropics' latest creation has also sparked an unusual wave of panic.

And that's because their mythos model is apparently fantastic at hacking computers.

According to the company, it's the equivalent of one of the best hackers in the world,

able to find critical flaws known as bugs in all sorts of software and systems that are vital

for modern life.

It is capable not just finding a bug,

but exploiting that bug,

going through all the links of the cyber kill chain

to successfully execute attack and doing so autonomously.

In fact, according to Anthropic,

it's too dangerous to release mythos to the general public. Politicians and

CEOs have taken note, and even MI5 have reportedly been called in to warn the UK's key

infrastructure companies about the risks. So, is this the dawn of a terrifying new era in cybercrime,

or a clever marketing move from one of the world's most powerful AI companies?

I'm The Guardian Science Editor Ian Sampal, and this is Science Weekly.

Isha Dow, you're a reporter covering artificial intelligence for The Guardian, and you've written about Anthropics mythos. Tell me a bit more about its apparent capabilities.

There's a lot we don't know. What we do know, based on a release from the company itself,

as that it's found a lot of what are called zero-day vulnerabilities,

...