John Gruber

What's that?

Are you doing it?

I'm going to do this.

We're just going to talk.

Let's talk about mythos.

So,

We haven't really talked about this on the show yet, but a little while ago, Anthropic released or sort of didn't release its new model, Mythos, which it said is so good at writing code that it's actually dangerous, that it found vulnerabilities in every operating system and browser, all kinds of stuff, and it launched this whole thing called Project Glasswing basically as a cybersecurity measure.

They were like, we've built AI that is too powerful.

We need to work on mitigating it together.

There's been this really fascinating backlash to it since then, where there are a lot of people all the way up to Sam Altman at OpenAI who were like, this is essentially scare tactics marketing.

They're just trying to frighten you.

None of this is real.

There are people who are going through and saying, well, actually, they're extrapolating the data out from a little bit of manual review of some things that they found old bugs that aren't even relevant in old software anymore, all the way to people being like,

This is the end of the world.

The spectrum has been wild.

And this week, a couple of interesting things happened.

One of which was that it came out that a group, we think associated with a vendor of Anthropix, got access to Mythos and has been playing around with it, as far as we know, without any terrifying goals behind it.

But they've been in there mucking around.

Mythos is also apparently potentially Anthropic's way back into deals with the US government.

We've talked about this a lot.