Sam Sabin

I'm Larry Mantle.

This week in AI has so much news for us to get to, and we're joined by Axios cybersecurity reporter Sam Sabin joining us.

Sam, thank you so much.

Let's start off with Anthropix reported breaches of government websites.

And a lot of these rogue agent incidents are really raising questions about, okay, how do we need to better adjust our defenses, right? So it accounts for the fact that An agent or a model is going to need like a rotating set of like permissions, right, to do their tasks.

And I'm struck by you saying it's like an employee going rogue because that's a whole different mindset to try and consider, well, what might a human being do? do that you wouldn't want them to do, because that is a challenge, working with humans to anticipate ways in which they could do something that you don't want them to do.

We're talking with Sam Sabin, cybersecurity reporter at Axios.

Also with us is AirTalk producer Nick Perez.