049 - How Do Attackers Use Local LLMs to Phish At Scale?

So, selfishly, I'm going to go first on this one.

And...

There's my first cough.

And I found this one particularly interesting because of the implication of what occurred with this particular case.

And, of course, there is my...

It's always sunny in Philadelphia reference for the episode.

But it's the news item that I chose for News React this week is the fact that with a flick of their fingers, the U.S. government seems to have.

prevented a new LLM model from being pushed to market, aka Anthropic and Fable slash Mythos, right? So Anthropic actually released Fable and Mythos to, well, Fable to the public, if you're not familiar with the two.

Fable is basically Mythos with a lot of guardrails on it, is essentially what it is.

And after they released it, Allegedly what happened, this came out after the fact, but allegedly what happened, Andy Jassy, Andrew Jassy, I don't remember, the CEO of Amazon, basically went to the U.S. government and said, hey, there's some national security concerns with this model.

The US government turned around and told Anthropic, hey, you can't let foreign nationals use this model.

And of course, Anthropic not being able to tell who was sitting at the computer using their model, it was just safer for them to just pull it from market, right? So we have a case now where the US government has basically shut down the offering of a private company, right? And I brought this one up because it kind of ties into the larger discussion that we've been having on and off on the show throughout several episodes about the power of local compute and not being dependent on workflows that live in the cloud, right? And imagine if this particular model, Fable, had been out for two months, three months, six months, whatever.

And then it was pulled from the market after people had had time to build workflows around it, build it into their day-to-day, and then all of a sudden it disappears forever.

Again, building the case for that local LLM capability, right? And of course, having the power to run LLMs, meaningful LLMs locally on site.

That's a whole different discussion.

I think we're going to have to be having another episode here on in a bit, Eric, about the availability of hardware because I have some gripes about that.

Anyway, that's my news react item for the episode.

What do you got for us, my friend?