KM

Kevin Mandia

Things Kevin Says on Podcasts

Where to Find Them

Kevin Mandia writes a16z Podcast .

Recently: “Building Defense for the Agentic Era: Kevin Mandia” on a16z Podcast (October 2026); “Building Defense for the Agentic Era: Kevin Mandia” on a16z Podcast (October 2026).

What They Said

“We created 20 full kill chains at Armadin that humans have done in the real world. … And we had no model go through the entire kill chains of more than eight. … And here's what's weird, by the way, we tested the open weight ones and the most advanced closed models. They all found eight.” — Kevin Mandia, a16z Podcast

A kill chain is the full sequence of steps in a successful intrusion; Mandia's company built a test from 20 that human operators had completed. He said the closed models were faster, but the open ones reached the same result when left to run longer. His conclusion was that the gap between open and closed models is smaller in cyber than in other fields.

a16z Podcast · 2026-10-06 Permalink → Listen →
a16z Podcast Around 23:21 into the episode
Speaker 1

…it's almost like a lunar landing race, right? The AI base. And you have R D people and they're doing the work to create models in a way where even those CEOs are like, we can't slow it. Let's get the government to help us slow it. That means you can't even control your own innovation. I have views

Speaker 2

on that, but we need to get another time. And so

Speaker 1

…time are made by the red teamers. You know what I mean? They're the ones that understand this stuff. And we created 20 full kill chains at Armadin that humans have done in the real world, period, at different victim sites and our experienced operators have done when testing networks. And we had no model go through the entire kill chains of more than eight. So that's where it was eight out of 20. And here's what's weird, by the way, we tested the open weight ones and the most advanced closed models. They all found eight. So if you're, yeah, so it was all about just speed and cost. And the closed models were faster to finding exploitable risk, but that's coming down. But we kind of let the open models run longer and they got to the same place. So in the cyber world, the differentiation between closed and open is not as great as in other domains, probably. And it's compressed. Yeah. Yeah. I would say that's…

Speaker 2

consistent in terms of like the capability gap at least closing a little bit. But that's interesting that their performance is basically the same.

Speaker 1

From my perspective, seeing the charts from the team, all the lines ended up in the same place. And when you're looking at, it was immediately time, cost, and then call it effectiveness or creativity. They all ended up at the end of all their operations where they hit diminishing returns, they…

Speaker 2

not on cost. Yeah, that makes sense. Yeah, that makes sense. So it's a decent segue maybe to talk about what kind of models you guys are using. And then what role do you think the lab companies play in the future?

Speaker names come from our own diarization. The position is estimated from where the line sits in the episode.
“The whole, let's slow down the models, we don't want cyber risk. Too late. The open models are already good enough and these things are common now.” — Kevin Mandia, a16z Podcast

Mandia was explaining how AI-led attacks differ from human ones in scale and speed. He argued that finding exploitable vulnerabilities is a structured problem that doesn't need the most advanced model. What holds criminals back for now, he said, is that they can't yet get GPU capacity anonymously.

a16z Podcast · 2026-10-06 Permalink → Listen →
a16z Podcast Around 06:09 into the episode
Speaker 1

…where it's at today. It'll get better and cleaner. But the differences are, first and foremost, the scale of what AI can do dwarfs humans, like in ways humans don't even get. So you have a scaling problem in that humans could always find only one path into a network. Yeah, they had to be selective

Speaker 2

because they had to devote their limited resources to one direct path, right?

Speaker 1

Yes. And then so scale is a challenge. Speed, ridiculous. What AI does in a microsecond would take 70 humans. They can't even do it. It's apples to oranges. And then, so what was always lacking is AI creative or effective. But when it comes to what we do, we don't need the fanciest model. We're not trying to speak 400 languages with our models and all that kind of thing. What Armadan's doing on offense is we're finding vulnerabilities, exploitable risk. That is code. That's a structured language, a structured process. Because it's structured, AI is going to be great at it. Right. Right. So I really think it's already here today. Like the whole, let's slow down the models. We don't want cyber risk. Too late. The open models are already good enough and these things are common now. It's just a matter of the minute you have anonymous availability of GPUs, you'll see far more criminal attacks. Oh, interesting. Yeah, you know what I mean?…

Speaker 2

Okay. So you, your experience working in the security industry for 30 years, you probably saw a fair amount of nation-state attacks, right? Every day. Every day. So talk about the differences or similarities between nation-state attacks. And I use that just to say the most sophisticated, most…

Speaker 1

So everything's going to change rapidly, right? But I can tell you nations on offense have never, in my opinion, they've never really been when you're hacking for espionage and for security reasons, you hack with what I would call kind of a sniper round. You're not spraying and praying. For the…

Speaker 2

yeah so then that is a good segue back to armadin so um you have talked about the defense needs to be a great offense right yes like like the best got to train your defense with something yeah of course you got to train your defense with something and then it needs to be continuous right right so…

Speaker names come from our own diarization. The position is estimated from where the line sits in the episode.

Collections They Appear In