podcasts / The a16z Show / episode
The a16z Show cover art
podcast
📅 2026-11-13 ⏱ 46 min episode · ~2 min read 📝 Transcript · 8,247 words 🎙 Ami Luttwak

Every Day Is Zero Day | Wiz CTO Ami Luttwak: The a16z Show transcript

The a16z Show · · Transcript & summary by WhipScribe

Wiz CTO Ami Luttwak explains how AI is compressing the time from vulnerability to attack, and what security teams should stop doing and build now.

Read the transcript Episode page ↗
In this episodeWizElon MuskOpenAIGoogleDeepMindMythosAtlasPHPLinuxWindows XPScan for GoodOWASP

▸Chapters

Tap a chapter to play it from that moment in the show's own player; the transcript opens there too.

Transcribe your links. Paste a YouTube, Google Drive or podcast link and WhipScribe gives you the transcript. Transcribe a link →

▸Notable quotes

“We are calling this Black October since it's, in my view, like a Black Swan event.” Ami Luttwak · at 1:36 — ▶ hear it from here
“AI attacks now are prevalent, easy, but also better than any human, okay?” Ami Luttwak · at 41:46 — ▶ hear it from here

▸What this episode covers

Ami Luttwak, CTO of Wiz, frames October as a turning point the hosts call Black October. He argues two shifts happened at once: AI models can now find and exploit complex flaws, and open source models have caught up to the frontier faster than expected. The result, he says, is attacks that compress months of expert work into about twenty minutes, from a small API mistake to stolen AI keys across customers. He also says the speed problem is worse than the flaw count, because details can leak before a patch exists.

The conversation then turns to what defenders can do. Luttwak says patching will not be solved by tools alone: startups built with automated development, testing and deployment can plan for auto-patching, but most legacy companies cannot. He recommends attacking your own systems continuously, starting with unauthenticated internet exposure before code scans, and running layered pipelines that move from cheap deterministic checks to expensive models so scanning budgets are not burned. He says much of what security teams do today is compliance and reporting rather than security.

Looking ahead, he says secure architecture, sandboxing, isolation and lockdown matter more than new spending, and he warns that legacy software in the real world, from refrigerators to traffic systems, is exposed. He expects breaches and worms to return in a pattern similar to the 1990s and 2000s, but with AI attackers better than humans at complex zero-day work. The hosts close by noting that the old playbook of delaying releases for security no longer fits.

▸Every Day Is Zero Day | Wiz CTO Ami Luttwak on The a16z Show: the answers

Who is Ami Luttwak?

Ami Luttwak

Ami Luttwak is the CTO of Wiz, the cloud security company that the episode says is now part of the Google family. He joins the a16z hosts to explain AI-driven attacks and how security teams should respond.

What did Ami Luttwak say about Black October?

Ami Luttwak said security professionals are calling October 'Black October' because it was, in his view, a Black Swan event. He said the month brought more security incidents and more severe vulnerabilities than anyone had seen in a single month, including a PHP flaw he called probably the most severe ever found by AI.

“We are calling this Black October since it's, in my view, like a Black Swan event.”Ami Luttwak · 1:36

What did Ami Luttwak say about PHP?

Ami Luttwak said almost 100% of Wiz's customers had some PHP running, and that about half had started patching within the first two or three days. He warned that because PHP runs on about 70% of the internet, even this single vulnerability will take years to fully resolve.

“but they're very, very aware of the importance of patching.”Ami Luttwak · 7:38

What did Ami Luttwak say about Red Agent?

Ami Luttwak described Wiz's Red Agent as the offensive side of its AI defense, which attacks a company's applications from outside using context from inside. He argued that the best defense is attack, because an application considered secure yesterday may not be secure once attackers get new model capabilities.

“What the red is trying to say is the best defense is attack, right?”Ami Luttwak · 21:57

What did Ami Luttwak say about Scan for Good?

Ami Luttwak said Scan for Good is an initiative from Wiz that offers to scan public services that want help and look for problems. He said it will not solve their issues, but he hopes the findings will push those organisations to start running their own scans.

“Scan for Good is trying to help any public service that wants help, and we try to scan them and try to find problems.”Ami Luttwak · 44:22

What is Ami Luttwak's main argument on The a16z Show?

Ami Luttwak argues that AI has made attacks faster and more capable than human attackers, so the old security playbook of periodic scans and static defenses no longer works. He says companies must attack their own systems continuously, automate detection and response, and design secure architecture from the start, because budgets alone will not solve the problem.

Mentioned in this episode

Episode facts

Released
November 13, 2026
Length
46 minutes

▸Key takeaways

01Wiz CTO Ami Luttwak calls October 2026 'Black October,' a Black Swan month with more security incidents and more severe flaws than he has ever seen.
02Attackers now move from an initial opening to full compromise in about 20 minutes, a process Ami Luttwak says took the best attackers weeks or months.
03Nearly all Wiz customers ran PHP during the PHP-to-shell flaw, and about 50% patched within two or three days, yet years of exposure remain.
04Luttwak says Linux fixes can take weeks or months to reach every distribution, while AI can exploit a flaw in minutes, so patch times are not falling.
05Luttwak's verdict: only companies that design for constant self-attack and automated response will survive, because AI now beats humans at zero-day attacks.

▸Questions this episode answers

Q1What did Ami Luttwak say about AI attacks during Black October?

Ami Luttwak, CTO of Wiz, called the month of October 'Black October' and said it was a Black Swan event, with more security incidents and more severe vulnerabilities than he had seen in any single month. He said AI attackers now operate at a speed and expertise no human attacker matches, and that open source models were catching up to frontier models faster than he had expected.

Q2What did Ami Luttwak say about the PHP vulnerability and patching?

Ami Luttwak said nearly all of Wiz's customers were running some PHP, and that about half had patched within the first two or three days, a fast drop. He warned that the problem will take years to clear, because PHP runs on roughly 70% of the internet and some older PHP versions no longer receive a patch at all.

Q3Why does Ami Luttwak say patching is still so slow?

Ami Luttwak said AI has made turning a patch into a working exploit much faster, but patches have not kept up. Information sometimes leaks before a CVE has a fix, and in Linux a reported kernel flaw can take weeks or months to reach every distribution. He said automated patching is realistic mainly for new startups with automated development, testing and deployment, not most legacy companies.

Q4What did Ami Luttwak say about Wiz's Red Agent and continuous testing?

Ami Luttwak described Red Agent, Wiz's AI attacker that tests a company's applications from outside using context from inside, and said the best defense is attack. He argued defenses that were secure last week may fail once new models arrive, so companies must attack themselves continuously. He said most Wiz customers already use it and that scanning unauthenticated internet exposure comes before code scanning.

Q5What does Ami Luttwak say security teams should stop doing?

Ami Luttwak said most of what security teams do is not security. He argued that much of their time goes to compliance and reports, and to generic configuration checks he called a list of configuration checks he does not care about. He said teams should focus on finding critical risk, mitigating it, adding detections and being ready to respond.

Q6What did Ami Luttwak say about security budgets versus secure design?

Ami Luttwak said security budgets will not solve the problem on their own, because secure design matters more: sandboxing, isolation, network and process lockdown, and assuming zero-day flaws in the Linux kernel will exist. He said budgets may rise and CISOs should review their vendor stack for AI-native tools, but focus and architecture come first.

Q7Why does Ami Luttwak say startups with AI keys are being targeted?

Ami Luttwak said attackers now target startups that hold AI API keys, and that an entire attack can run in about 20 minutes, from entry to taking the AI keys of all customers. He advised founders to set spending limits on AI keys so stolen keys do not generate large charges, and to invest in security now.

Q8How does Ami Luttwak compare the next few years to the 1990s?

Ami Luttwak predicted the next two to three years could look like the 1990s and 2000s, with remote, unauthenticated exploits and breaches appearing all the time. He said he would not be surprised to see worms return, as in the Windows XP days, and argued AI is now better than humans at complex zero-day attacks, which was not true then.

▸Transcript

Our own transcript of the public episode, with a timestamp about every 90 seconds. Tap a timestamp to play the episode from there in the show's own player; the link button on a paragraph copies a link that opens this page there.

0:00We're going back to the 90s and 2000s again. We're going to see breaches popping out all the time again now. It's going to look very different than the last 15 years of security. AI attackers that are actually going after companies, startups in scale and speed that we've never seen. Elon Musk tweeted that recently, every day is zero day. What is AI doing to make this happen? Models are good enough now to find and exploit complex vulnerabilities. And the other is that the open source models, that's the scary bit. A few months ago, we said we have a few months until open source models catch up. So now they're catching up. The security game has just changed. You need to be thinking about how you can automate so much of both the blue side and the red side

0:41right now. Most of what security teams do is not security. We should stop doing most of what we do and focus on finding critical risk, figuring out how we mitigate it, or at least put detections in and also response readiness for it. What do you think the next several years looks like? AI attacks are prevalent, easy, but also better than any human. That's the problem. The only companies that will be able to survive are companies that are... on both sides. Ami explains how attackers are using AI to find vulnerabilities and move from an initial opening to a full compromise at machine speed, while patches still move slowly through the software ecosystem. They also discuss what defenders can do, continuously test their own systems,

1:36automate investigation and response, and build infrastructure assuming vulnerabilities will exist. The question now is whether defenders can adapt as quickly as attackers. Ami, thank you so much for joining us. Security professionals are calling October Black October, and it's clear that we're entering a new era of agentic attacks. Can you, just to start off, walk us through what the heck is going on right now and how do we get through this critical moment? Yes. So first of all, great to be here. We are calling this Black October since it's, in my view, like a Black Swan event. We've never seen, never, ever, seen so many security incidents happen in a single month. And we are talking about multiple different types of security incidents, both the number

2:29of vulnerabilities that has been found in October and the severity of those vulnerabilities, including the PHP vulnerability, probably the most severe vulnerability ever found by AI, right? So vulnerabilities coming in a rate that we've never seen. But on the other hand, even worse, we're seeing the rise of AI attacks. So AI attackers that are using completely new speed expertise that we've never seen before, actually going after companies, startups, refrigerators in scale and speed that we've never seen. Yeah, I mean, the scary thing about that, to your point, is, and you guys cover some of this in your blog as well, the speed. And I think you've even said, definitely to us privately, but probably publicly as well, that every day is zero day, right?

3:15Elon Musk tweeted that recently. What is AI doing to make this happen? What is the mechanism that it's able to speed writing exploits? I think you guys actually did some work around this where you found it collapsed the timing from months to weeks. Say more about how that works. And I think your point on the audience for this podcast, what should founders be thinking about and doing? So if we think how we got here, right? So there are two major moments in TBL October. One was the mythos moment where Mythos came out and said, oh my God, we can find vulnerabilities. That was one moment. But again, nothing happened. We said, oh my God, it's coming, but we still had to... People say, okay, okay, the sun is still rising.

3:59The other big moment that we had was the Hacking Face incident with OpenAI. So that was another major incident where we saw the risk of AI attacks, but again, it was accidental attacks. What happened now in Black October is that all of those potential theories, the things that we said are coming, they all came and crashed upon us and it's happening right now, right? So one is that the models are good enough now to find and exploit complex vulnerabilities, right? So, and that means that this is not a theory anymore. It's happening. It's an explosion of vulnerabilities. And the other is that the open source models, right? That's the scary bit. And we said it when we spoke about METOS a few months ago.

4:43We said we have a few months until open source models catch up. Okay? So now they're catching up. It means that now it's not just about vulnerabilities. They're able to find ways to run code in your environment. It doesn't have to be a complex PHP vulnerability. It can be just a mistake in your API, right? They can find and utilize those zero days in your code, right? And basically run an entire attack that usually took for the best attackers weeks or months in, again, around 20 minutes. That's what we're seeing now. So your entire... How far behind are the open source models? I mean, when we're chatting with folks on the offensive capability side, it tends to be about like six to nine months or so.

5:28But that's close enough now at this point. But I'm curious what you've seen. There's two different questions. One is, OK, on the offensive side, right, to find a vulnerability, are we better than the open source models? Yeah, maybe we have three months, not more than three months, right? The problem is that is the offensive side on attacking, okay? Because I'm a company. It doesn't matter. Let's say it's a startup, okay? And again, we are talking about a wave of attacks. If we just look at AI development providers, right? We had a blog about, we've seen initially the blog had nine companies attacked in October and the list is growing every day. We're now on 15. So like, it's just an companies attacked in October, and the list is growing every day. We're now at 15.

6:05Wow. Like, it's just an example of one market, right? Like, AI development platforms, okay? These are getting targeted specifically now because they have AI keys, right? The fact that an attacker, it doesn't have to be a sophisticated attacker, can use AI. Open source models are good enough to find problems in your code. And the big problem is to also use them in a matter of 20 minutes. So think about which startup is ready for detecting those attacks. I don't know. And then actually finding them and responding to them in time, right? Like it's you need a 24-7 soak in every small startup. Yeah, right. Why don't we talk for a second on PHP to Shell as an example of just a case study of one of the many that came out during Black October here.

6:54Can you talk a little bit about just how prevalent you saw this across your customer base? But then now that we're actually a couple weeks out, I'm really curious your take on how has patching actually been across the customer base and what are kind of the state of things a few weeks out now that we've had a few weeks of seeing it be out there? So PHP is maybe one of the most prevalent technologies as an external facing right? Oh yeah. As a former CISO of a PHP e-commerce company. Yes. I think it's like 70% of the internet or something. Yeah, exactly. You know, the impact is basically almost 100% of our customers had some PHP running, right? So we saw a major drop very quickly. So customers are now, I wouldn't say they've automated patching yet,

7:38but they're very, very aware of the importance of patching. So I'm talking about 50% patching within the first two, three days, right? So like almost immediate reduction, but still the problem is that this is 70% of the internet. So even just this single vulnerability is going to take years, okay, until it's resolved. And there is many, many, many cases that we see both on the internet and also the customer environments that they don't have any way to patch. Okay. Because other versions of PHP are not even getting a patch. Yeah. Can I ask just a meta question here? And two of our partners wrote this great piece on time for writing exploits versus patching. And there's a lot of dialogue out there about how AI has made it just cheaper, easier to turn a patch diff into a working exploit.

8:30And there's sort of some data on just sort of that time falling. On the other hand, it seems actually that the data showing time to patch actually hasn't gone down. In fact, it's gone up if you look at the data over the last 20, 25 years. Can AI help with that side as well, Ami? Or is it still early days? So first of all, you're right. The time to get an exploit dramatically went down. It's even worse because many times the information leaks before the CV even is out. Okay. Many times even the CV is out, the patch is not out. Okay. Many times, even the CV is out, the patch is not out. Okay. In the PHP case, we worked for months with PHP to get the entire industry ready and it leaked before time.

9:12Right. So the customer doesn't have anything to do. The patch is not ready. In Linux right now, even if there is a vulnerability reported to Linux kernel, it might take weeks or even months until it reaches all of the distros. Right. it might take weeks or even months until it reaches all of the distros, right? So the industry in general is not ready for it. Even if you can patch internally quickly, sometimes until a patch gets to your distribution, okay, it's not ready for AI, right? AI can exploit things in minutes. So the answer is that the world is not ready. The world is not ready for this. It's a problem. In terms of patching, you're right. We don't yet exactly know how to automatically patch.

9:50And I honestly don't see a path towards that. It's easy to say, yes, let's automate everything. In reality, if you are building now a new company, you can do it, right? If you're building a new company and you have automated engineering and automated CICD and automated testing, you can basically be at a position where I tell you, patch this and your entire system can automatically do it. And I think for new companies, you should design your system to be able to auto-patch, right? But in order to do that, you need an energetic pipeline for everything, right? You have automated dev, so you need the ability to take a ticket and have an automated dev to create the patch. Fine. Then you need automated testing,

10:32automated ICD, and automated DevOps. There's not a lot of legacy companies that have all of that, and it's not even realistic for most of their code. But for startups, maybe it's possible, and I would consider this a task for startups. If you start, plan this, you know? Yeah, absolutely. I'm curious to just dive in real quick on the, a little bit deeper on the offensive side. So going through kind of the discovery of this, it was funny, a little bit of a footnote here, but this was a vulnerability that was independently discovered by two groups, Wiz and Caliph, around kind of the details of the PHP to shell vuln, which is a thing that used to happen a bunch back in the day. I'm curious, though, for your research team, how were they really leveraging AI in terms of discovering

11:23this and just going a little bit deeper on how it's really unlocking vulnerability research teams? So a few things. First of all, I think this will actually happen much, much, much more now. Okay. Because every time there is a new model, there's new capabilities. Yeah. Again, I look at it like Bitcoin mining, okay? Yeah. Again, I look at it like Bitcoin mining, okay? Yeah. Number of vulnerabilities will go down, but because it's like Bitcoin, there's always more. And the new models can uncover more vulnerabilities, right? So it's not surprising that there is a new model, okay? And then multiple teams find new vulnerabilities, right? Exactly, yeah. And by the way, right now, we are getting ready for ZTC,

12:09the zero-day competition of WIDs. And we're seeing, we're getting multiple, multiple reports on a very, very critical vulnerability that is also, I believe, is going to be a major impact on the internet. And this vulnerability is being reported by multiple teams. Yeah. It's amazing how it brings everything old is kind of new again in the security industry right now. Yes, you're right. Internally, our vulnerability team, we built basically also for the Wizz product, we built what we call Atlas, which is a harness for vulnerability detection, right? So we're using that with different models all the time because every model has its own unique capabilities from OpenAI or Google or Carbon. Each of them has its own capabilities.

13:00We're trying to use all of them and find whatever we can. But the point is that everyone is doing it. So you would expect a lot of teams to find vulnerabilities at the same time. And that's the important thing. We have to focus on infrastructure that is critical, using the models of the frontier AI that have a few months of advantage. And we have to continuously do that on those areas to make sure that we find the B4 attackers. But there's so much code in the world that it doesn't help because there will still be vulnerabilities that attackers find on an area that we didn't look into. I don't want to take us down a frontier model rabbit hole too much. But just to probe a little bit on what you said

13:45about using multiple models and each one being good at different things. To the extent you can share, Ami, I'm curious about two points there. One is what are different models better at, at a high level? And then two, how do you see, I mean, you know, obviously, Wiz is a part of the Google family now. And so you, how do you see, I mean, you know, obviously, uh, Wiz, um, is a part of the Google family now. Um, and so you have, uh, the DeepMind team and their models at your fingertips, but it sounds like you are using multiple models. Um, how do you think about the sort of relationship between Frontier Labs and independent cyber companies, um, in this day and age? So I think that every model and also every version of every model, right?

14:27And now it's getting complicated because you have the frontier models and you have the cyber versions of those models. It's a complicated matrix, right? But the point is that each of them does behave differently and has a different speed and cost, right? So there is no single answer. That's my point. Like, if you want to do a task, it's for cyber, but it's true for other stuff as well. The only thing you can do as an independent company is to try all the models all the time and to find the best combination, right? And the way I see it is that, you know, people say, oh, now it's easy to develop using AI. It's not true. For most founders right now, the key is going to be the cost, the cost of the service, right?

15:16And the only way to create a scalable service is by creating a very, very intelligent pipeline where you do some of the tasks with specific models and some of the more complicated tasks with other models and you mesh all of them together in a very, very efficient way. Otherwise, you're going to lose money on your sales, right? So as an example, right? We have what we call the blue agent which investigates the tax, right? I cannot run the most advanced, expensive model on every event, right? But I can create a pipeline where you start from a deterministic detection. Then you go through a cheap model detection. Then you go through a more advanced model. And if all of them think it's malicious,

15:59I will say to the most advanced model to determine is it really malicious before I tell the customer, you see? Yep. So you need multiple models, multiple types of models, and you have to design pipelines. And I think that is really the new engineering challenge for founders, for companies building AI systems at scale. Yeah, absolutely. And just on this cost point, you talk about in, I think, a blog post that you guys wrote, but maybe summarize a little bit here, this paradigm shift that needs to happen around continuous testing, right? Sort of the focus being on the word continuous versus static. And cost obviously plays a big role in, you know, sort of the democratization of every company being able to do that, not just the sort of most profitable and sophisticated.

16:51How far are we from getting there on a cost, from a cost perspective? And what else, I guess, needs to happen? Maybe it's also cultural shift, et cetera, like, you know, tooling shift. But how far are we from, I think, where do you think companies need to go and how far are we today? So, I mean, first of all, it's a big question. Like, let's take security, for example, and go back to the Mintos moment, right? So people say, oh my God, there's Mintos. I have to scan all of my code, okay? Fine, let's say you did it in April. Okay. You might have wasted $5 million. Okay. These are real numbers that you could have wasted a lot of money. It's very expensive. Now, okay. Now there's me just 1.2. You do it again.

17:36Like, this is not, like, this is, I'm not talking about small budgets. I'm talking about, some customers are spending now tens of millions of dollars on like scanning code. Like you have to think about a scalable system, right? Continuous system that scans, but it's intelligent. You can't just, so because there's going to be new models, new capabilities, new code, right? So we see this as a major challenge for the industry. We're trying to figure it out. And I'm not going to tell you that we solved it, right? But the challenge is, how do I use the budget wisely in a way that scans from the inside, from the outside? We have active attacks, we have passive attacks, we have code analysis, we can scan on the disk itself,

18:18we can look at the runtime. How do I use all of the budget in the most efficient way for the customer? Not as a single scan, but as a continuous scan, right? And that is the challenge of security scanning right now. If I tell you, oh, yeah, just run Meet us on your code, great, you're right, you can do that. But does that actually solve your problem continuously? No, it's just going to get you like a big hole in the budget. I know it feels like a lifetime ago that we all used to talk about just cloud security, pre-AI security. But I'm curious, you know, look, for security teams and practitioners, CISOs, security engineers, all of this, just because new problems appear doesn't mean old problems go away.

18:57And so for a lot of, you know, Fortune 500, Global 2000, they're still dealing with a lot of cloud security challenges. I'm curious if we, for the first part of this question, let's just kind of set AI aside. And can you talk about kind of the state of cloud adoption and security in 2026? And then for the second part of the question, let's layer AI in there and talk about how it's really changing cloud security specifically. Yeah, so, I mean, first of all, cloud security, when we talk about I have an application that's running, it's always in the cloud, right? When we talk about AI attackers coming after my application, they will always start from the same vulnerabilities and problems and mistakes that you could have done before.

19:45Right? So, like, nothing changes in the fact that in cloud security, you have to make sure that you're focused on critical risk and then you fix it. Right? Yeah. If you have a PHP vulnerability you didn't fix, yeah, attackers are going to come from there. Right? So, like, we just added more vulnerabilities and more types of attacks. It doesn't mean you don't have to fix the old ways of attacks, right? So, in terms of cloud security, what we're seeing is that, first of all, everyone understands that you need this system of record that understands and gives you visibility into everything, right? So, I would say what we're starting to see change is, of would say what we are starting to see change is of course

20:25that companies are trying to create more agentic workflows and also expect more from Wizz, right? So, for example, if before we would provide a vulnerability with prioritization, now they would expect us to go back to the code, find the code, create the fix, create the fix, create the pull request, you know, so basically help automate more of the processes that take time. And that's also true for detections, right? They don't want just a detection. They expect also to get a full investigation, forensics, find the root cause, tell me how the attacker got in and push them out, you know? So the way I see it is that what AI is doing right now is basically every step that we had that we didn't solve with Wizz automatically,

21:12companies are trying to do it automatically on their own and they also expect us to get better, right? So as an example, we are not finding threats anymore. We have to do a full investigation. We also need to do forensics. And I think the next step for us, now that we see AI attackers, we need to start thinking about automated response, which is very scary. Yeah, absolutely. Yeah, completely agree. Yeah, maybe on the topic of things that you guys are automating and just how Wiz has evolved, you talked about Blue Agent. Can you share a little bit more about Red Agent? Because I think that's where, you know, you're really simulating an adversarial point of view. And, you know, is your response,

21:57is Wiz's response to AI shifting risk, you know, essentially, are you giving your customers, going back to this sort of speed thing, like speed of the defender, speed of the attacker, are you sort of thinking of ways to give machine speed to your customers, if you will? Exactly. So there's eventually there's two forces here, the red and the blue. The blue tries to find attacks and stop them. What the red is trying to say is the best defense is attack, right? So you have to attack yourself with AI continuously because unlike what we had before where we had static defenses, now defenses have to be dynamic because every new model has new capabilities, right? So maybe your application was considered secure yesterday, but it's not secure today.

22:52Think about it. If I try to give you an analogy, think about you have a door, okay? And maybe you hide your key under the mat, okay? And that's considered secure, right? But your attackers are getting superpowers every week, right? So maybe next week, they're going to be able to get into your door. So whatever was considered secure last week, if every AI model can get into the door, it's not secure anymore. So the watermark keeps going up, right? So the only way to know you have a problem is to continuously attack yourself with the latest models. So before the attackers can open the door, you figure it out, you fix it, and you don't have a lot of time. And that's the speed, we'll talk about speed of defense, right?

23:36It's crucial. So the red agent tries to do that, tries to do an offensive AI attack against your applications from the outside, using the context we have from the inside to try to find the problems before the attackers actually use them. Are you doing a similar pipeline on the red team side? So my question is coming from the angle of teams who this all makes sense to them. They want to get started with it. Let's help them avoid a bunch of the mistakes that everyone else has had to learn. On the blue team side, you start this with kind of deterministic approaches and then you layer in cheap models and the pipeline that you talked about progressively getting more expensive. Are you doing similar approaches

24:14on the red team side as well, where it's you're starting deterministic and then you're layering cheap models? Or are you layering just a whole bunch of different models, all starting from scratch and all running simultaneously? Yeah, everything. Yeah, everything. Yeah, exactly. So, yes, I mean, first of all, let's compare this to, like, if you have a budget, you can just give it to Mithos and run on code. That's not very efficient. So more efficient is that you analyze all the exposure, right? That's what we do at Deterministic LeanWiz, right? Then we layer on top of that what we call the exposure rate agent that just looks at everything that is unauthenticated, exposed on the internet from your environment, right?

24:55So that's an immediate analysis for everything, every exposure you have. That is the most urgent thing to do. It's even more important than scanning your code, okay? Who cares about your code? I'm talking about the actual unauthenticated exposed tax surface that you have. More urgent, right? Next step is to start doing deeper pen test analysis and deeper code analysis, right? So there's definitely layers here. And also in terms of models and investigation time, from the outside, what we call exporter agent, maybe it will take a few minutes for every endpoint because you have a million endpoints. When there's an investigation time. From the outside, what we call exporter agent, maybe it will take a few minutes for every endpoint

25:26because you have a million endpoints. When there's a critical application, I might take 10 hours to try and hack it, right? So it's all a pipeline. Everything is going to be a pipeline now, right? You cannot throw money on code scanning and think you're done. No, it's all about managing the budget. This is the whole point, right? And they're all connected because if I find a problem and the engineering says, oh my God, this is too expensive to fix. Okay, you know what? Let me tell the blue agent to put some detections there. You see, this is all working together. Right, right. I was going to ask, and it's really interesting to hear your point of view on this, because we had an internal discussion with Merck the other day about how it feels like state of the art is essentially what Red Agent is doing.

26:13You know, you have agent swarms constantly attacking your own systems so you know how to protect against them, right? You're working with some of the top enterprise customers, right? Over 50% of the Fortune 100. And then you also have leading edge startups in your customer base. How many people are actually doing this? So in with a majority of customers are already using the red agent. I think across the industry, not enough. I think everyone should do it right now. Like there is nothing more important than to scan your environment continuously from the outside. Like there's nothing more important than that. Like the best defense is attack. When they come, it might be too late, okay? So really the best thing you can do

26:58is continuously attack your environment. And so in my view, like again, on the Wizz customer side, it's like we are getting very quickly to almost full coverage right like everyone understands it and it's really one of the fastest growing features that we've ever released and in the world I don't think people understand I think there's sometimes you know me too got a lot of PR good okay but the PR went to the wrong place. I actually don't think the scanning code is the most efficient thing you can do, okay? And it's important to understand that because people can say, oh, yeah, I invested so much in scanning code. Really, that's not what you should do. You should focus on scanning your external exposure before

27:37because, like, we had a, there was a big incident just two days ago. A key hard-coded in a mobile application that exposed a lot of people's data, right? That's where attackers get their attacks, unauthenticated surfaces. That's what you have to run scans for and you have to do it continuously. So I think that's the gap and the world needs to know that. Yeah. Maybe on this thread of what founders should be doing. So Zane and our colleague Joel actually wrote a note in the midst of Black October, you know, to our founders talking about spending more on security, potentially considering delaying product releases. So I want, this is a question for both of you, maybe Zane to say a little bit more about that.

28:23But say I'm a founder with a release coming up. I have a backlog of security issues. How should I think about what needs to happen before I ship? So question for both of you, I guess. Yeah, sure. I mean, I can kick off. I mean, I think you and I are very aligned on this, which is the security game has just changed, right? And you need to be thinking about how you can automate so much of both the blue side and the red side right now and really leveraging a variety of agents for those automations. And so I think historically in the old models, it was, you know, you had a release and then you kind of delayed that for security and all of that. I think that the new model that you're thinking about

29:04is how can you be doing all these things in parallel? How can you be really leveraging a variety of agents to be continuously testing? And if you're building a new company today, really doing patching from the beginning and building in so much automation into your software development and your infrastructure development. Yes, yes. And again, I don't want to scare, but let's just say I work with the CEOs of startups that are getting attacked. Let's just be in their shoes, okay? You wake up and your security team says something bad happened overnight. We tried to stop it, but they got in. We woke up after an hour, and that's in the good situation. Yeah, you're being very nice there of an hour.

29:52They even have a security team, okay? After like a round, I don't even know if they have someone that is dedicated, right? And the attacker, yeah. And we found out that the attacker got in and the entire attack, again, takes 20 minutes until, from the time they get in, until they take all of the secrets, all of the AI keys of all of the customers, right? And now you have to figure out what are you going to say to all of the customers and what you're going to do. And this is what is going through companies right now. So the point is that AI attacks, we've never seen that before. We have to understand it's a new threat, right? It's a threat to everyone. But if I'm a startup, especially if I have anything attackers want, again, especially

30:36like if I have AI keys or anything related to something they can use, right? I am on their target. It doesn't have to be a big company, okay? I am on their target. It doesn't have to be a big company, okay? Startups are actually their key target now for LLM checking, right? What do you need to do? If you have AI keys, probably better to put a limit on them because if they steal them, you don't want to get a lot of charges, which we see happen a lot. It's really, really sad. But also make sure that if you have a target on your back, you do need to put a lot of charges, which we see happen a lot. It's really, really sad. But also make sure that if you have a target on your back, you do need to put a lot of focus on security.

31:09This was never the case, but it's now true that if they want to get in, it will happen fast. So you do have to invest in this. There's no other way to put it. Yeah, I think that's exactly correct. The level of security investment, whether that is dollars, time, any of these things just needs to go up the next next year, which is, you know, we all anticipated that it would go up, but that's probably on the higher end. I'm curious, what, you know, what are you seeing folks spend more money on? And on the flip side, what should CISOs be spending less money on, right? And, you know, you've kind of written publicly about a class of security products that may become somewhat obsolete, But maybe talk through that in terms of what,

32:10you know, where budgets should go up, but also where they should come down. Okay. So first of all, again, let's assume you have a house, okay? And you know that an elite group of the most trained hackers are going to hack your house, okay? They might not come through the door. They might come through the window or the roof, right? Honestly, how many houses are even ready for that? Like most houses would say, fine, I'm not going to secure the roof. Obviously, there's no way for me to, I mean, it's impossible, right? So the point is that like we are really, most of the companies are not ready for attackers that have the speed and expertise of the best attackers in the world. And this is why it's a hard problem.

32:54So even before we talk about budgets, I don't even know if budgets are going to solve the problem. OK, it's not about security budgets. It is about focus. But to be honest, like it's about security architecture, security design, like it's a budget is not even going to solve anything. Like it's, it's not like there is a security tool in the world you can buy and everything is fine. Dang it. I'm not just, you know. Would have made a lot of our jobs a lot easier if that was true. But yeah, yeah. I don't think it's true. Obviously, most of the latest security tools are not even relevant anymore, okay? But even if you get security tools that is AI native, that understands how to attack attacks in AI work

33:37and cloud attacks and so on, even then, to be honest, it's about security design even more than just security budget. Okay? So is security budget going up? Maybe. Should we review the existing stack that we have? Yes. Okay? But more than that, it's about having a dev team that is focused on scanning the code and applications all the time. It's about thinking about sandboxes and isolation and secure design and network lockdown and process lockdown and image lockdown. Like, it's just, like, if you're building stuff now, get ready for a nation state attacker. You see, like, get ready for that. Imagine there's going to be a vulnerability. Like, don't assume, ah, let's run everything in a Kubernetes cluster.

34:23No, like, there's zero-day vulnerabilities in the Linux kernel all the time. Anyone can get out of a container. So you see, like, secure design is much more than just the budget. Obviously, security budgets, that's great, but it's not going to solve it for me. Yeah, that's a great point. To that point, it definitely feels a lot like cloud in the sense of this is a real stair step where I'm seeing so many CISOs take a look at their existing vendor stack and really reboot that to see what are the AI native ones. as you are really focused on secure architecture is exactly what you're talking about, Ami, of like really thinking about the problem through and understanding there are vulnerabilities

35:08at every stage of the stack there. Simultaneously, this is the old CISO saying of like, don't let a good crisis go to waste, which is absolutely happening here of rebooting so much of your security product landscape to actually be AI native, to help you go into this new era that you already are in, whether you want to be or not. Yeah. Well, maybe just to get explicit, not to dunk on any particular company, but what's a type of, you know, software, let's say security software that you have in place, that CISOs have in place now that they just don't, no longer need? Okay. Most, I've always been against, again, it might be a harsh sentence, okay? Most of what security teams do is not security, okay?

36:02There may be small, small subset of their time is actually security. And the way to check it is this. And that's why I always believe in just attack yourself. How much time are you spending with red teams and how much time you're spending with compliance and creating reports, right? To be honest, most of the security teams are not focused on security. And security, in my view, is getting either, what you're doing is either getting ready for the attacker, attacking yourself, or putting guardrails for the attacker, attacking yourself, or putting guardrails for the attack when it comes. Okay? So the point is that we do have to be much more efficient now on security teams. We should stop doing most of what we do and focus on

36:35finding critical risk, figuring out how we mitigate it, or at least put detections in and also response readiness for it. We have to be good so we don't have time to spend on everything that is not focused on that. So if you do a task, you say, is this report really helping us, you know? And that was always my point against like classic, what we call CSPM in cloud, which was just the way you did cloud security before. Just a list of, yeah, configuration checks. I don't care about that. You have to care about where is the attacker going to come from? Am I doing everything I can to stop them? Can I, if there's a strategic barrier I can build, let's build it, you know? Like focus on understanding

37:20the actual real risk in the environment, creating a sustainable long-term isolation strategy. These are the things you should do. Yeah. And actually on this point of time management, and maybe flipping a little bit to how AI is actually helping security teams, right? So as agents maybe take on more of the routine investigations and testing, how do you think, and you've answered this a little bit, but maybe deep dive into it, some of the most experienced security professionals should be spending their time? Much easier for a security team starting from now from scratch. Okay? Like, obviously, if you're building a company right now, you're lucky and you should try to build everything fully automated.

38:05Security view is automated. Patch process is automated. Investigation is automated. If you have enough context and knowledge base about your environment, there is definitely a way for you to build everything completely automated. The problem is that for most security teams, it's just not the case, okay? Not the case. You have a lot of legacy software, thousands of developers. You don't even know who owns what. It's like, it's not like everything is changing because there's an AI agent. Okay? And it still goes back to find the context of who actually owns every resource. How do I get them to patch stuff? It's complicated. So it's going to take time for most of the security team in the industry

38:43to figure out how to start really, really getting to efficiency. It's not as easy and simple as it sounds because a lot of the knowledge of the organization is a bit, you know, obscure. It's not anywhere that is actually written. But it is the challenge of all security teams. If I cannot respond in minutes, what can I, am I relevant? Yeah. So to that point, I'll kind of round us out with a question here on connecting the through line of what we've talked about, which is what do you think the next several years looks like? Like our next two to three years. Like I'll start you off with my own prediction, which is that we're going back to the 90s and 2000s again. This is, there's going to be full remotes,

39:32unauthenticated, full ODAs popping out all the time now. We're going to see breaches popping out all the time again now. It's going to look very different than the last 15 years of security. I think we are going back to a era that looked very different in security. And I'll even throw one wild one out there that like, I wouldn't be surprised if we end up seeing worms again on the internet like we did in the Windows XP days. We were chatting internally on this on the infra team and Martine brought up a funny point of there was for all of us who were kind of in broader tech and it at the time there was a window of time where you couldn't install and update windows xp fast enough because from when you installed it when you started

40:18the update a worm would hit your system and infect it during the time it took to actually update the system and so you literally had to it took to actually update the system. And so you literally had to install all of this offline because if you installed it online, it would get compromised faster than you could patch it. And I think so much of this that feels so new with AI in a lot of ways is something plenty of us experienced 20, 25 years ago in the kind of wild west days of the internet. And so building off of that, I'm curious what you think the next few years look like. Ami, just to add a quick question to tag on to Zane's, you wrote this great blog post in April that there was a Y2K moment coming in 12 to 18 months.

40:57To Zane's point on the next two or three years, would you pull up that moment now, given everything that's happened in the last six months? Yeah. So first of all, I think Black October might be the beginning. We didn't want to scare too much and we were hoping there's more time, but the open source models are catching much quicker than we thought. The hope was that open source models are a year behind, but they're not. I think it's happening. We've never seen this level of attacks across customers. It's happening. And if you want to compare to the 90s, our dependency on technology is much, much, much higher. Oh, absolutely. And also what happened, and it's important to understand, this is almost the only area in humankind right now

41:46that AI is better than humans, okay? That's the problem. And that was not the problem in the 90s. AI attacks now are prevalent, easy, but also better than any human, okay? That's the problem, okay? I'm talking about attacks that are very, very complex, using zero days. We had today an attack using a zero day on a kernel driver. My God, like that's very, it should be very rare. It's happening every day now. Human attackers cannot do it. So you see, like we're at an era where AI is better than humans. And if I need to, I cannot foresee two years ahead from now. I think the only companies that will be able to survive are companies that are building now, not just security automation, security design, security reviews. They're designing their systems

42:34that are exposed to the internet in a new way. And whatever is not ready should take it, take off the internet. Okay. Quickly. Right. So on the internet, you need to be ready with everything that is like ready for the crazy attackers. And you have to design for this, right? I think that's the only way. A lot of this podcast was focused on, you know, enterprises, tech companies, et cetera. I want to bring up Scan for Good, which is an initiative you guys launched recently. But I want to talk about in the context of some of the attacks, you know, we're talking about AI infrastructure companies and things like that. But, you know, if we look at the recent attacks on refrigeration units in Europe, which, you know, if you think about the food supply chain, et cetera, right, like this at a mass scale has dire consequences for humanity.

43:29And I don't want to fearmonger, but, you know, I think there's also this appropriate line that you're walking on me of making sure people are prepared and are taking things seriously. So I guess two questions. One is how worried should we be about what's happening to public infrastructure and maybe these entities that don't have huge cyber budgets, etc.? And two, maybe talk a little bit about Scan for Good and why you guys are doing that, how you're, you know, helping nonprofits, public services, et cetera, get really prepared for what is already happening. So I, I, I mean, yeah, I am worried. Um, uh, I mean, those exactly, I mean, the legacy, uh, software, it can be refrigerators. I mean, that are actually in the real world. Uh, it can be, um, uh, traffic. I mean, it could be anything.

44:22It can be really anything that we know that is connected to the internet is at risk. That's exactly the reason we started Scan for Good. Scan for Good is trying to help any public service that wants help, and we try to scan them and try to find problems. It's not going to solve all of their issues, but we're trying to give them a bit of a push. Our hope is that if we find stuff, then they will understand it's a bigger project and they will start doing red scanning on their own. Right. So it's like this is more getting them to push into that. This is awesome. Thank you so much. This was incredible. Really great to see you both. And yeah, I mean, this is a crazy time. Oh, absolutely. Crazy time. Thank you for what

45:03you're doing right now. Absolutely. Who would have thought from OWASP conferences to here, buddy? Good to see you. Yes, yes. I remember. It was a fun one. Thanks for listening to this episode of the A16Z Podcast. If you liked this episode, be sure to like, comment, subscribe, leave us a rating or review, and share it with your friends and family. For more episodes, go to YouTube, Apple Podcasts, and Spotify. Follow us on X and A16Z and subscribe to our Substack at a16z.substack.com. Thanks again for listening and I'll see you in the next episode. As a reminder, the content here is for informational purposes only. It should not be taken as legal business, tax, or or investment advice or be used to evaluate any investment or security

45:47and is not directed at any investors or potential investors in any A16Z fund. Please note that A16Z and its affiliates may also maintain investments in the companies discussed in this podcast. For more details, including a link to our investments, please see a16z.com forward slash disclosures.

▸More episode summaries

The a16z Show cover art
The a16z ShowFrom the Telephone to Modern Computing | James Gleick2026-11-14 · 62 min
The a16z Show cover art
The a16z ShowThe Next Era of Games: Mods, Agents, and Infinite Worlds2026-11-12 · 45 min
The a16z Show cover art
The a16z ShowAlex Imas & Tyler Cowen | AI Is Moving Fast. Why Isn’t the Economy?2026-11-16 · 64 min
Transcribe your links. Paste a YouTube, Google Drive or podcast link and WhipScribe gives you the transcript. Transcribe a link →
New The a16z Show transcripts by email
One email per new episode, unsubscribe in every message. Not shared with the publisher.