How I Invest with David Weisburd

Sanaz Yashar on AI accelerating cyberattacks and the shift to preemptive defense

51 min Listen to the original Episode page
CybersecurityArtificial IntelligenceZero-Day VulnerabilitiesRisk ManagementInsuranceEnterprise Security

Summary

Sanaz Yashar argues that AI is accelerating the discovery and weaponization of zero-day vulnerabilities, making attacks faster and more accessible. She highlights that while AI offers defensive potential, widespread adoption is hindered by trust issues, high costs, and the 'human in the loop' requirement. Yashar notes that AI agents attack using multi-step chains rather than isolated vulnerabilities, necessitating a shift from patch-and-scan to surgical prevention. She predicts mass security breaches when AI enables new actors to cause physical or critical infrastructure damage. The discussion also covers the economic impact of cyber risk, including insurance assessments and the stress on CISOs.

Key takeaways

  1. 1
    Companies face a 'double/triple' cost burden when using AI for security: paying for AI tokens, paying human staff to operate/verify the AI, and paying for security infrastructure.

    Unidentified speaker (speaker_0) · 15:47

Episode sections

  1. 0:00 – 4:51

    AI Security Risks and the Zero-Day Arms Race

    Sanaz Yashar argues that AI is accelerating the discovery and weaponization of zero-day vulnerabilities, making attacks faster and more accessible to non-state actors. She notes that while AI offers defensive potential, widespread enterprise adoption is hindered by trust issues and the high cost of implementation.

  2. 4:51 – 8:46

    The Mythos Stress Test and Vulnerability Prioritization

    Yashar discusses Anthropic's Mythos release, noting that while the intent was good, businesses struggle to fix the massive volume of discovered vulnerabilities. She highlights that only a tiny fraction of discovered issues are actually exploitable, making blanket patching inefficient.

  3. 8:46 – 15:04

    Attack Chains vs. Siloed Vulnerabilities

    The speaker explains that AI agents attack using multi-step chains rather than isolated vulnerabilities. Therefore, fixing 99% of minor issues is insufficient if a single low-severity flaw completes an attack chain, necessitating a shift from patch-and-scan to surgical prevention.

  4. 15:04 – 20:02

    Barriers to AI Defense Adoption

    Yashar identifies three main barriers to AI-driven security: economic costs (double/triple spending on tokens and humans), psychological resistance to change, and cultural inertia in workforces untrained to use AI efficiently.

  5. 20:02 – 27:55

    Collateral Damage and the 'Chernobyl' Event

    Yashar predicts that mass security breaches will occur when AI enables new, rule-breaking actors to cause physical or critical infrastructure damage. She cites the recent PHP vulnerability as a near-miss event that demonstrated the potential for widespread disruption.

  6. 27:55 – 31:37

    Preemptive Defense and Kill Switches

    To counter unpredictable AI agents, Yashar advocates for preemptive defense strategies that simulate attack chains and implement 'kill switches' before breaches occur, rather than reacting after an alert is triggered.

  7. 31:37 – 33:45

    Legacy Security Tools and Vulnerability Remediation

    Speaker_0 discusses how customers have varying resources for fixing vulnerabilities, noting that legacy systems do not scale for AI-era threats. They outline a three-step process: gaining visibility into new zero-days, identifying exposed attack chains, and determining mitigations using existing tools.

  8. 33:45 – 37:11

    Scale of Vulnerabilities and AI Adoption

    Speaker_0 cites a Fortune 2-digit retail company with 36 million vulnerabilities and a Bank of America report requiring 1,000 hires to fix issues from one release. They note that while 90% of vulnerabilities can be mitigated by existing controls, the remaining 10% require AI-driven autonomous workflows or agentic remediation.

  9. 37:11 – 40:25

    Insurance and Economic Impact of Cyber Risk

    Speaker_1 and Speaker_0 discuss how insurance companies are now assessing security before setting premiums, indicating a lack of confidence in corporate self-assessment. They debate whether the economic cost of cyberattacks will be absorbed by companies, financial markets via insurance, or government subsidies.

  10. 40:25 – 42:05

    Zero-Day Frequency and CISO Stress

    Speaker_0 states that zero-day vulnerabilities now occur daily, a significant increase from once a week or year. They share an anecdote about a Morgan Stanley CISO who noted that their job security depends on no major vulnerabilities occurring, highlighting the existential stress of the role.

  11. 42:05 – 50:42

    Collaboration and Workforce Dynamics

    Speaker_0 observes increased collaboration among US companies and startups, contrasting it with the siloed nature of the US market compared to Israel. They note that junior security personnel are becoming more effective.

Claims worth verifying

It took a whole year to find a zero-day and a long period to make an exploit kit during Yashar's time in the military.

Unidentified speaker (speaker_0) ·2:19

How to verify: This is a personal anecdote; verify by checking Yashar's LinkedIn profile for her tenure at Unit 8200 and cross-referencing historical cybersecurity case studies from that era to confirm typical zero-day discovery timelines. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

Out of all new discoveries by Mythos or other frontier AI labs, only 1 out of 1.3 million are actually exploitable.

Unidentified speaker (UNKNOWN) ·5:54

How to verify: Check Anthropic's public reports or blog posts regarding the Mythos release and any subsequent data on vulnerability discovery rates versus exploitable findings. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

Autonomous car projects started in 2009 and took eleven years for humanity to feel safe enough to release them.

Unidentified speaker (speaker_0) ·6:45

How to verify: Research the history of autonomous vehicle development, specifically looking for the start date of major projects (e.g., Google's self-driving car project began in 2009) and the timeline for public road testing approvals. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

Mythos was released in April of this year (2026).

Unidentified speaker (speaker_0) ·7:39

How to verify: Check Anthropic's official website or press releases for the exact launch date of the 'Mythos' product or stress-testing initiative. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

Zafran Security has a small language model called ZLM trained to find vulnerabilities and deploy fixes.

Unidentified speaker (speaker_0) ·11:13

How to verify: Search Zafran Security's official website, whitepapers, or technical blogs for documentation on their 'ZLM' (Zafran Language Model) and its capabilities. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

A top financial customer tested LLMs/SLMs for employees for two months and found people used expensive modules for simple tasks.

Unidentified speaker (UNKNOWN) ·17:34

How to verify: This is an anonymous case study; verify by searching for public case studies from Zafran Security or similar cybersecurity firms detailing internal AI usage audits. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

The recent PHP vulnerability affected 70% of internet-facing websites.

Unidentified speaker (speaker_0) ·25:35

How to verify: Search for the specific PHP vulnerability referenced (likely a major CVE from late 2026) and check statistics from security firms like W3Techs or Sucuri regarding the prevalence of the affected PHP versions. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

Nation states follow unwritten rules in cyber warfare, such as avoiding actions that cause physical harm like crashing planes.

Unidentified speaker (UNKNOWN) ·21:46

How to verify: Review academic literature or reports from cybersecurity think tanks (e.g., Carnegie Endowment, RAND Corporation) on international norms and deterrence strategies in cyber warfare. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

Binance suffered a $6 billion hack.

Unidentified speaker (speaker_1) ·1:51

How to verify: Search news archives and Binance's official statements for reports on a $6 billion security incident. Note: Yashar clarifies this was a system transition attack, not a direct crypto theft, which should be verified against primary sources. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

Do your own research

  • Verify Anthropic's Mythos release notes and public testing data to confirm the 1 in 1.3 million exploitability rate.
  • Check recent CVE databases for the PHP vulnerability mentioned and its impact on internet-facing websites.
  • Review Bank of America reports on cybersecurity hiring needs and remediation timelines.
  • Analyze insurance industry reports on cyber risk premiums and pre-underwriting security assessments.
  • Examine Zafran Security's technical documentation for their custom small language model (ZLM) capabilities.
  • Assess Fortune 2-digit retail companies' public disclosures on vulnerability remediation resources.
  • Look for CISO job security metrics or industry surveys on stress levels related to zero-day frequency.
Full transcriptexpand

0:00Unidentified speaker (UNKNOWN): The

0:00Unidentified speaker (speaker_0): nation would control AI will control the world. And there is now the race to AI, right? It doesn't matter if AI developed in US or developed in China. It's like martial arts, right? It can be used against you if you are not having the right policy and the guardrails and if you don't control it.

0:19Unidentified speaker (speaker_1): We haven't seen this mass security breaches that people are predicting. What's the timeline looking at? that and is it inevitable?

0:29Unidentified speaker (speaker_0): I want to say that I have an answer to that, but I don't have a full answer to that. I think it's going to get much worse before it gets better. We are going to have big incidents in cybersecurity caused by threat actors weaponized with AI or by AI agents.

0:46Unidentified speaker (speaker_1): One of the issues and dangers of AI is that now a single actor, somebody in their basement that might just diabolical and just not care about even their own life could get a hold of AI.

1:01Unidentified speaker (UNKNOWN): How

1:01Unidentified speaker (speaker_1): do you deal with something like that?

1:03Unidentified speaker (speaker_0): Prevention, prevention, prevention. Understanding since you don't have almost control over intention anymore

1:10Unidentified speaker (UNKNOWN): in

1:11Unidentified speaker (speaker_0): the time of AI and then the operation is fast and scaled. You need to understand what are the possible paths that this threat actor will harm you. So you need

1:23Unidentified speaker (UNKNOWN): so

1:23Unidentified speaker (speaker_1): Sanaz, you studied how attackers infiltrated systems at Unit 8200, which is a famous Israeli military command. You're also at FireEye, and now you're at Zafran, Zaffran, which is backed by Sequoia, uh Cyberstarts and Mem. Menlo. At Zaffran, you're defending some of the biggest and most valuable companies in the world. Recently we had this attack on Binance by some accounts, six billion dollar hack

1:55Unidentified speaker (UNKNOWN): of

1:56Unidentified speaker (speaker_1): of of the crypto firm.

1:58Unidentified speaker (UNKNOWN): What

1:58Unidentified speaker (speaker_1): do you make of that?

1:59Unidentified speaker (speaker_0): It's very interesting because uh that hack actually. was not a cryptocurrency hack. That was a cyber attack on the systems than transitioning that cryptocurrency. There was years uh when I was um working for military that it would take us a whole year, David, to find a zero day. And um it would take us like a very

2:25Unidentified speaker (UNKNOWN): uh

2:25Unidentified speaker (speaker_0): big period of time. long period of time also to make an exploit kit, to weaponize that zero day vulnerability in order to actually hack. Today, AI scanners finding zero days every day. And there is like exploit kit, like that weaponized vulnerability is existing before the fix comes out, right? It's because of the AI. And we are going to see more and more of this. You would ask, like, okay, so why we don't use AI also to defend ourselves,

2:57Unidentified speaker (UNKNOWN): right? Like

2:59Unidentified speaker (speaker_0): it's pretty forward. If we use if they use AI to attack us, why we don't use AI to defend us? And I think uh the answer of that is um we don't yet trust completely AI. Like there are like amazing demos and amazing companies

3:18Unidentified speaker (UNKNOWN): um

3:19Unidentified speaker (speaker_0): coming out of cell. But between the demo and fully implementation of AI for a very big bank to rely on security and to do a genetic fix, we still have a lot of

3:35Unidentified speaker (UNKNOWN): uh

3:35Unidentified speaker (speaker_0): steps. One of these is um

3:39Unidentified speaker (UNKNOWN): you

3:39Unidentified speaker (speaker_0): cannot cash AI if you need like to put a human as a gateway. If you cannot trust AI, because AI can get rock, right? I just told you. So, how can you get the full benefits of AI, right? So it's like a full circle. And since we are talking to investors, I would say AI. economy cannot scale without developing security. Like it we can invest as much as we want into AI and we create this bubble, but the moment security is not going step uh by step, even like earlier than how a AI is being developed, it's not going to be implemented and then we won't see the result. results of it uh in the market.

4:29Unidentified speaker (speaker_1): This episode is presented by Juniper Square, the

4:31Unidentified speaker (speaker_2): operations partner for private markets.

4:35Unidentified speaker (speaker_1): There's no question that cybersecurity is gonna be one of the hottest sectors in the next next five years. I think no one is questioning that. A lot of people are questioning how will the rest of AI businesses function w with with such a looming issue around security? Roughly six months ago, Anthropic released Mythos to the business community first in order to stress test their systems. What do you think about this approach? And is that where we see future LLMs

5:08Unidentified speaker (UNKNOWN): kind of

5:09Unidentified speaker (speaker_1): going forward?

5:09Unidentified speaker (speaker_0): I think the intention was good. Like definitely it was uh a preemptive protection act, right? Like uh we want to release it to a very few amount of um

5:25Unidentified speaker (UNKNOWN): uh

5:25Unidentified speaker (speaker_0): business uh partners in order to create the fix before the mythos is completely out. Um what realistically happened is the fix was not developed as fast as uh we wanted because uh of the trust, because of the scale, because of the volume, and because of the business. The business cannot just slow down the business and we say, Okay, now we fixed six million vulnerabilities. The numbers that I tell

5:54Unidentified speaker (UNKNOWN): you,

5:54Unidentified speaker (speaker_0): they are right. And the other piece of that was that out of all the new discoveries by Mythos or by other frontier AI. labs, just one out of 1.3 million are actually exploitable. Then the question is why a business leader should fix 1 million vulnerabilities in one application when this when this application is bringing like 30% of the revenue for this bank. So people started to understand that uh not everything that AI discovered. discovers actually is hackable and

6:29Unidentified speaker (UNKNOWN): not

6:30Unidentified speaker (speaker_0): everything is hackable is actually fixable that fast.

6:35Unidentified speaker (UNKNOWN): I'm

6:36Unidentified speaker (speaker_0): going to d give you an example in comparison for that. You know, 2020, for the first time people started to drive autonomous cars. The project started in 2009. It took eleven years

6:51Unidentified speaker (UNKNOWN): and

6:51Unidentified speaker (speaker_0): I would say it was like so slow, right?

6:54Unidentified speaker (UNKNOWN): Um,

6:56Unidentified speaker (speaker_0): for the humanity to feel safe enough to release

6:59Unidentified speaker (UNKNOWN): uh

7:00Unidentified speaker (speaker_0): that kind of autonomous system out. Why? Because we were afraid of accidents and we were going to die. So I think AI is not different than that. Like we just talked about the collateral hack. Unfortunately, I'm I wanna want be to very optimistic. in this uh in this episode because I do believe that at the end AI is becoming utility and we will have like guard rays around it. But today uh we are not there yet and and we can cause more damage and and more harm than value. So if uh for autonomous cars took eleven years, now imagine

7:36Unidentified speaker (UNKNOWN): what

7:36Unidentified speaker (speaker_0): does it mean? Like Mythos was released April this year and today is already like Fable Five and others that is like Mythos kind are available for the public.

7:48Unidentified speaker (UNKNOWN): And

7:48Unidentified speaker (speaker_0): if I tell you that it was enough time

7:51Unidentified speaker (UNKNOWN): for

7:51Unidentified speaker (speaker_0): the public to actually create those raw guardrails around it, it was absolutely not. So at this moment that we are speaking of we have more risk uh um being um exposed because of AI than actually defense.

8:09Unidentified speaker (speaker_1): I love the autonomous systems. I got my Tesla Model Y, and I pretty much use it 95% of the time. One of the reasons why I took 11 years was because there's so many edge cases. Even in midtown Manhattan, And sometimes my autonomous car will just stop at the intersection because there's too many pedestrians that have to take over because it's not ready for that use case of so many pedestrians. Is AI security in in the same vein? Is there a Pareto principle where you could solve 99% of vulnerabilities relatively quickly and then you're really focused on that 1%? Or is it just a different paradigm?

8:46Unidentified speaker (speaker_0): For now, yes. And it's not just about one vulnerability.

8:49Unidentified speaker (UNKNOWN): It's

8:50Unidentified speaker (speaker_0): a little bit more. complicated. Agents, um, whether if there is a human behind it or whether they are completely working autonomous, they don't think in silos. They think in uh actual chains. What does it mean? Today's security NAI AI is very kind of siloed, right? You have a protection for identity, you have a protection for fishing, phishing, you have a protection. for um vulnerability. But when AI starts like to actually find a way to bypass, or when a thread actor, a North Korean thread actor or Chinese redactor try uh to use it, they actually use all the different weaknesses in all the different systems and make one chain that makes AI um successful to infiltrate. and to do the mission. That means you can patch update 99% of the vulnerabilities and stay with one vulnerability that is like low

9:52and not critical, but still in a concept context of a chain, it will become the next hugging phase. face. So the idea is

10:04Unidentified speaker (UNKNOWN): not

10:05Unidentified speaker (speaker_0): just fixing as much as you can. I'm actually

10:10Unidentified speaker (UNKNOWN): like

10:10Unidentified speaker (speaker_0): there's different ascolas not in the security. I'm one of them that I think the time of patch and scan all of it is completely over. It's about time that you do it surgical and you do it like very smart. What does it mean? Despite AI. being created by human, it's not thinking like human anymore. So now we need to use AI to simulate what are those paths, those chains he can find in order to hack our system. The moment you find those chains, still you cannot fix them all. The smart people will find that one node that collapsed the whole chain. and go fix that

10:52Unidentified speaker (UNKNOWN): one.

10:53Unidentified speaker (speaker_0): And this is why I think the security should get to validation, validating like AI agents and those paths, and then to prevention. Look, the AI time is different than the time that we are living in. Like every day there is new improvement. I can tell you the systems and the algorithms and the models. that we are using. We also built our own model, a small language model, we call it ZLM, it's the Afron language model, that uh being trained to find vulnerabilities and to actually deploy the right fix. I I would say that's six months ago we were not there, David. It's just six months. Like companies that were created, six months. years ago, they cannot speed up uh this way of the model. So when everything is fast, faster than optic fiber, waiting for an alert to tell you something's wrong is too late. It's already too late. So you need to bring the whole protection and defense uh prior to that alert.

12:02Unidentified speaker (UNKNOWN): And

12:02Unidentified speaker (speaker_0): I would call it you should go from managing vulnerabilities and weaknesses to preventing those attack paths.

12:11Unidentified speaker (speaker_1): You brought up such a good point, which is that AI agents will attack systems in completely non-human ways. The best analogy I could think about is in chess or even Alpha Go is that you would start to see they started showing these moves, these chess grandmeisters, these moves that the AI was doing, and it was just completely, it made no sense for like 10 moves. It was completely orthogonal to what a real chess, a human chess player. And that's actually how they catch cheating today, is they could see, well, is this like a human?

12:44Unidentified speaker (UNKNOWN): It doesn't seem like

12:45Unidentified speaker (speaker_0): human.

12:45Unidentified speaker (UNKNOWN): Or is it that

12:46Unidentified speaker (speaker_1): is there's this a is this a computer? And it's important. an paradigm because there's this catch twenty two where if you're not using AI to defend and if you're not trusting AI, a human being isn't isn't able to defend it because because it's almost like two different ways of thinking about

13:05Unidentified speaker (speaker_0): it our neuron systems is completely different uh from the way AI works. Who would the nation would control AI will control the world. And there is now the race to AI, right? And this is now we are seeing,

13:20Unidentified speaker (UNKNOWN): um,

13:21Unidentified speaker (speaker_0): as an example.

13:22Unidentified speaker (UNKNOWN): Uh,

13:23Unidentified speaker (speaker_0): 2026 uh Entropic released uh Mythos, right? What I'm trying to say,

13:30Unidentified speaker (UNKNOWN): it

13:31Unidentified speaker (speaker_0): doesn't matter if AI developed in the US or developed in China, it's like martial arts, right? It's like your ch your own. chess you said. It can be used against you if you are not having the right policy and the guardrail uh guardrails and if you don't control it, at least now.

13:50Unidentified speaker (UNKNOWN): Um I

13:51Unidentified speaker (speaker_0): don't claim that I would understand what's going to happen in ten years and how the wars are going to look like. Maybe it's actually going to be better. Maybe people won't be killed. It's going to be all in the AI a world. world. But uh but today

14:05Unidentified speaker (UNKNOWN): uh

14:05Unidentified speaker (speaker_0): there is a very, very intense uh AI war behind the scene. I would say before that it was just cyber war, but it's just AI war. Also race over the infrastructure, the controls, the implementations, the signatures, and uh therefore those frontier AI leadership. in the world has now more responsibility, right? Because if they don't put enough uh in one place they want to be completely completely agnostic. But in the other place we can see that

14:40Unidentified speaker (UNKNOWN): um

14:41Unidentified speaker (speaker_0): enemies of freedom of the Western world, they can use them against us.

14:45Unidentified speaker (speaker_1): Earlier in the conversation you brought up this point where AI needs to defend against AI. What's keeping that from happening? Is this a technological issue or is it a psychological issue where human beings just have an issue of letting go and letting AI fight for them?

15:04Unidentified speaker (speaker_0): Today it's less uh tech issue, technical issue and technology issue.

15:09Unidentified speaker (UNKNOWN): Uh

15:09Unidentified speaker (speaker_0): I would say it's a combination of three business and economy. The second would be psychology and the third would be culture.

15:21Unidentified speaker (UNKNOWN): Um

15:22Unidentified speaker (speaker_0): let's tackle on the business first. It's expensive today. I know the prices are going down, but

15:29Unidentified speaker (speaker_1): To defend yourself.

15:30Unidentified speaker (speaker_0): To defend yourself with AI, right? It's it's expensive. Like for each company to develop a full uh agency system.

15:38Unidentified speaker (UNKNOWN): uh

15:39Unidentified speaker (speaker_0): to defend themselves against AI while still

15:43Unidentified speaker (UNKNOWN): uh

15:44Unidentified speaker (speaker_0): they utilize headcounts also

15:47Unidentified speaker (UNKNOWN): to

15:47Unidentified speaker (speaker_0): use AI because it's human in loop it's very except expensive so so they pay double and triple uh for the tokens and for the people that operating it and for the security and still the concept is okay go uh patch it all and in order to patch it all it's slowing down the business you cannot slow down the business security cannot be business disabler it's the other way right so first is economy it's expensive and

16:18Unidentified speaker (UNKNOWN): uh

16:19Unidentified speaker (speaker_0): somehow still have business disruption in it for the security. The second piece is psychology and people cannot trust people. It's change. It changed for us. It's very hard. There was a research that they made people to

16:34Unidentified speaker (UNKNOWN): uh

16:34Unidentified speaker (speaker_0): brush their teeth the different ways that they brush. After a month people could not live with themselves and they wanted to go back uh to the same way that they were brushing. So every change is different. Wait, they when they created, I I think I read about this when they the electricity was invented, people thought that electricity is going to use to kill us all.

16:57Unidentified speaker (UNKNOWN): And it

16:57Unidentified speaker (speaker_0): wasn't. Yeah, it was protesting in the streets and stuff. So that culture,

17:03Unidentified speaker (UNKNOWN): and

17:04Unidentified speaker (speaker_0): that is the psychology, and the end of it is the culture. Today, a um workforce is not yet trained

17:19Unidentified speaker (UNKNOWN): to

17:20Unidentified speaker (speaker_0): use AI in an efficient and safe way. One of our customers that is a top financial customer, they did a test and they opened all the LLMs, all the SLMs for all the employees for two months

17:34Unidentified speaker (UNKNOWN): to

17:35Unidentified speaker (speaker_0): see what's going to happen, right? And uh after two months they saw the results, and of course, people used the most expensive module of LLMs for doing very simple tasks that they didn't want, used AI. And there was like very hard tasks that people didn't use AI at all, right? So the question is how we can safely, because there is an impact of this, you know, I'm a cyber. person. I was cyber before cyber was invented. But when I'm thinking of this, I think there are going to be roles that are going to be completely disappeared little by little because people cannot think in silos when they

18:18Unidentified speaker (UNKNOWN): uh

18:18Unidentified speaker (speaker_0): work with AI. And people that won't do that uh transition, they are going to find themselves out of the workforce, right? So that cultural thing that you worked with somebody, especially in the US, here people are working a lot of times in the same company. In Israel it's not like the same. We get bored and we move after two, three years. Here they work like 30 years in some company. And suddenly they need to be on boarded with agents and they don't have that capability. So culture-wise, you want to uh protect your people

18:55Unidentified speaker (UNKNOWN): and

18:55Unidentified speaker (speaker_0): you don't adopt AI as fast as possible.

18:58Unidentified speaker (speaker_2): One thing I've learned from talking to hundreds of investors is that great investment firms aren't built on investment returns alone. The firms under are great at the things most people don't see: their operations, their relationship with LPs, GPs and the quality of information they use to make decisions.

19:12Unidentified speaker (UNKNOWN): And

19:12Unidentified speaker (speaker_2): here's what AI has changed. Every firm now has access to the very same models, so the intelligence isn't the edge anymore. The edge is what you could feed it. A firm with its fund operations and data in one connected record can actually put AI to work. A firm running on disconnected systems simply can't. That's why thousands of GPs run their funds on Juniper JuniperSquare. Square. Juniper Square puts your fund operations, data, and administration together in one connected record. That means less time managing disconnected systems and more time investing, working with LPs, and building your firm. This episode is brought to you by JuniperSquare, the operations partner for private market GPs. Learn more at junipersquare.com slash how I invest. That's junipersquare.com/slash how I Find invest. us.

19:57Unidentified speaker (speaker_1): Maybe from a level set standpoint. Because

20:02Unidentified speaker (UNKNOWN): it's

20:02Unidentified speaker (speaker_1): in the zeitgeist right now that everyone's gonna be dealing with these security issues every day, that there's new security breaches, but we haven't seen this mass security breaches that people are predicting. What's the timeline looking at that? And is it inevitable? And is it the only way that we could make change?

20:26Unidentified speaker (speaker_0): I want to say that I have an answer to

20:28Unidentified speaker (UNKNOWN): that,

20:28Unidentified speaker (speaker_0): but I don't have a full answer to that. Um

20:33Unidentified speaker (UNKNOWN): I

20:33Unidentified speaker (speaker_0): think it's going to get much worse before it gets better. I cannot put a timeline timeline on that, right? And in my previous life I needed to put timeline on everything and it was like depending on the lives of people.

20:45Unidentified speaker (UNKNOWN): But

20:46Unidentified speaker (speaker_0): lately the time is becoming very liquid. thing, right? Like stuff that we are doing during the last eight months is not something that I would imagine that we would achieve in eight years. So

21:01Unidentified speaker (UNKNOWN): I

21:01Unidentified speaker (speaker_0): don't know exactly when it's going to happen, but in the near future, let's say, and near future can be six months, can be a year, but can be also five years. We are going to have big incidents. in cybersecurity caused

21:19Unidentified speaker (UNKNOWN): uh

21:19Unidentified speaker (speaker_0): by threat actors weaponized with AI or by AI agents, collateral hack itself. And sometimes

21:27Unidentified speaker (UNKNOWN): uh

21:27Unidentified speaker (speaker_0): the combination of that, and I tell you why I'm talking about this. When nation states are fighting each other in cyber, there is this unwritten rules, let's say this. like if Russia trying to hack all the time

21:46Unidentified speaker (UNKNOWN): uh

21:46Unidentified speaker (speaker_0): the defenses of US and other countries,

21:51Unidentified speaker (UNKNOWN): uh

21:52Unidentified speaker (speaker_0): it's in the rule. But if Russian threat actors uh cause a plane to crash, it's not in the rules. This is not

22:01Unidentified speaker (UNKNOWN): something

22:02Unidentified speaker (speaker_0): nations nation states would do with each with each other. It's

22:05Unidentified speaker (UNKNOWN): like

22:07Unidentified speaker (speaker_1): they wouldn't want somebody reciprocating that way.

22:09Unidentified speaker (speaker_0): Yes. It's like we call it like the red button and the pink button, right? You wanna you wanna have that deterrence that you can do it, but you won't press the red button. Like you won't crash planes uh by cyber because you cannot anticipate

22:26Unidentified speaker (speaker_1): And that's from game military game theory. You don't want to start an escalatory situation that you can't control.

22:33Unidentified speaker (speaker_0): True. Also you're kind of aware of uh others' capabilities in cyber. And you know that you cannot also defend against what's coming, right? And not eventually if

22:45Unidentified speaker (UNKNOWN): um

22:46Unidentified speaker (speaker_0): somebody being uh attacked by some cyber they can like kind of react in physical words. So that was like a kind of written rules between the nation states. I won't numerate any all the nation states because there are going to be some defend offended out

23:01Unidentified speaker (UNKNOWN): there.

23:01Unidentified speaker (speaker_0): Um lately, because of AI, we are seeing the bar is being lower. And this new countries that they didn't know enough about technology and they are not part of

23:16Unidentified speaker (UNKNOWN): that uh

23:17Unidentified speaker (speaker_0): apparatus starting to use AI

23:21Unidentified speaker (UNKNOWN): uh

23:21Unidentified speaker (speaker_0): to hack stuff. This is dangerous because it's so fast and they are not part of that doctrine of

23:29Unidentified speaker (UNKNOWN): what

23:30Unidentified speaker (speaker_0): I should do with AI and what I can do with AI.

23:33Unidentified speaker (speaker_1): And these are just smaller countries.

23:34Unidentified speaker (speaker_0): Smaller, smaller. I I don't want to name them but we are seeing like a lot of newcomers to the cyber wars, right? And they are growing fast.

23:44Unidentified speaker (speaker_1): It's like they're not part of this old boys' club of nation states hacking each other. They don't know the rules of the game.

23:50Unidentified speaker (speaker_0): One hundred percent.

23:52Unidentified speaker (speaker_1): This is what Eric Schmidt calls the Chernobyl event. He thinks that a lot of governments and also a lot of companies will not take AI security seriously until there's a significant situation of collateral damage. Maybe six billion dollars Binance hack is not it, but at some point there's gonna be this one event that creates so much mass casualties, whether financial or human casualties, that everyone will have to recalibrate their AI security. It'll be this thing that everyone is pretending doesn't exist. to something that comes to the forefront.

24:28Unidentified speaker (speaker_0): I think you're right. I think as much as every event that we find our private data exposed to the internet,

24:37Unidentified speaker (UNKNOWN): we

24:37Unidentified speaker (speaker_0): can tell ourselves, okay, maybe we are not important enough. Maybe it's not going to be developed to something like that much dangerous. I think the moment some routine works cannot be done by humans because of a hack, it's going to be

24:56Unidentified speaker (UNKNOWN): uh

24:56Unidentified speaker (speaker_0): the moment that people lose patience, right? The moment you cannot. We were almost there. Uh PHP happened

25:03Unidentified speaker (UNKNOWN): a

25:04Unidentified speaker (speaker_0): couple of days ago. There was a very wide spread of vulnerability being exploited, being found and exploited by AI.

25:12Unidentified speaker (UNKNOWN): And

25:12Unidentified speaker (speaker_0): the funny or sad thing is that they try to do it very responsibly, disclosure, but it was leaked. And then attackers start the game as the defenders to weaponize. And this is this PHP is in every website. It's in 70% of the websites that are internet facing in the world. So we were very close to see that. there is no banks, there is no post office, there is no education, like everything would be

25:45Unidentified speaker (UNKNOWN): just

25:45Unidentified speaker (speaker_1): the websites or the

25:46Unidentified speaker (UNKNOWN): It

25:46Unidentified speaker (speaker_0): was like everything that is internet facing. Now imagine that everything that is internet facing and you need it like to to work with something wouldn't work, right?

25:55Unidentified speaker (speaker_1): So this isn't you can no longer log into TikTok and watch TikToks. This is you go into your bank account and

26:02Unidentified speaker (UNKNOWN): it's

26:02Unidentified speaker (speaker_0): zero

26:02Unidentified speaker (UNKNOWN): balance.

26:03Unidentified speaker (speaker_0): It's down. It's down. Like it's down. And there was like actually a couple of hours that I checked a couple of banks and they didn't have service. Uh it was in the weekend and it was during the night, so maybe a lot of people didn't like feel it until it was like fixed. But next time it's going to be much more impactful. Um

26:24Unidentified speaker (UNKNOWN): And

26:24Unidentified speaker (speaker_0): just imagine phone systems being w one won't work, right? And we cannot communicate or traffic lights uh will go bad because

26:33Unidentified speaker (UNKNOWN): uh

26:34Unidentified speaker (speaker_0): because of that. And then bad things. Like I'm not sure if it's going to be Chernobyl, churnable. Uh like the nuclear systems that I also have some history with. I think it's going to be a couple of very annoying and very impactful chain chained um events that are going to be the limit. And then we are going to stop.

27:02Unidentified speaker (UNKNOWN): I

27:02Unidentified speaker (speaker_0): don't know if we can stop then. I think it's too late. Unlike nuclear stuff that you can kind of monitor.

27:09Unidentified speaker (speaker_1): Going back to what you said earlier, which is there's this old boys club of nation states. that have this, I guess, this culture of how to attack each other for lack of a better word. One of the things that's assuring about that is mutually assured destruction. A lot of the people running these organizations are high up in government. A lot of them are rational actors, they care about their lives, they care about the their leadership lives. One of the issues in and dangers of AI is that now a single actor, somebody in their basement that just might just be diabolical and just not care about even their own life could get a hold of AI.

27:53Unidentified speaker (UNKNOWN): How

27:53Unidentified speaker (speaker_1): do you deal with something like that?

27:55Unidentified speaker (speaker_0): Prevention, prevention, prevention, understanding since you don't have almost control over intention anymore

28:02Unidentified speaker (UNKNOWN): in the

28:02Unidentified speaker (speaker_0): time of AI, and then the operation is fast and scaled, you need to understand what are the possible paths that this threat actor will harm

28:15Unidentified speaker (UNKNOWN): you.

28:15Unidentified speaker (speaker_0): So you need kind of almost to be offensive to yourself before they come in, and then you need to fix those junctions. that can uh collapse as much as risk paths for them. I give you an example, like to defending your own house. You can put cameras in your house, you can put fences, you can put even a security guard out there. But

28:41Unidentified speaker (UNKNOWN): in

28:41Unidentified speaker (speaker_0): the world of AI, it wouldn't work, right? Now you have to be cameras, put puts cameras, you have to understand like what are all the different ways that will finish into. your house and to put cameras and fences and maybe guards um in the junctions before that are leading to your house. So that twenty nine minutes is definitely not enough in to contain a certain hack. That means you should be much more preemptive and preventive, right? In advance. You should uh have visibility how uh agents backed by humans and undergrounds or not undergrounds are going to use those agents and and to make sure that uh you have guardrails and you have kill switches for those chains.

29:32Unidentified speaker (speaker_1): And this is the these chaining of vulnerabilities, meaning you get in through one entrance and then you have to get through through these multiple kind of linear entrances and you you just can't afford to wait until they're in the house.

29:57Unidentified speaker (speaker_0): severe but also with um a data uh server or data container that has uh clear passwords and non-encrypted stuff in it and they need just one to find in order to uh find that chain and they do it so fast and so unpredictable And that uh you cannot predict it by human eyes. Like I would say that even if you were the best hacker from the NSA or from

30:30Unidentified speaker (UNKNOWN): um from

30:31Unidentified speaker (speaker_0): A two hundred, you would still feel it feel it's complicated. It's that the chest thing you said, right? To uh predict all the different ways

30:41Unidentified speaker (UNKNOWN): uh

30:41Unidentified speaker (speaker_0): that an agent will find himself in. So you need to use agents, like now we have good agents and bad agents, right? Fairy agents and evil agents to kind of predict what the bad agents can do and put uh limits and fire bricks almost

31:00Unidentified speaker (UNKNOWN): in

31:00Unidentified speaker (speaker_0): place.

31:01Unidentified speaker (speaker_1): So now that we've sufficiently scared everybody, there are people like Zafron working on solutions. Yes.

31:08Unidentified speaker (UNKNOWN): You

31:08Unidentified speaker (speaker_1): have some of the top Fortune four to 500 five hundred companies going to you across multiple different domains, banking. infrastructure, and everything. So a company comes to you and says, we want to keep ourselves from from getting hacked. What's the framework on how

31:27Unidentified speaker (UNKNOWN): you

31:27Unidentified speaker (speaker_1): approach that?

31:28Unidentified speaker (speaker_0): First, um we need to look at their environment and see how they are doing it today. Um because the initial point which each of these customers are different, right? We have customers that they have like 11 or 12 teams to remediate one vulnerability. I'm not kidding what I'm saying now, right? And we have customers that they have some internal systems they built to kind of bring context from uh other uh teams together. So they did some work. So we need to understand. what is the initiation and how much does it take them today to fix vulnerability. And then how much does it cost

32:11Unidentified speaker (UNKNOWN): them

32:13Unidentified speaker (speaker_0): to take down

32:15Unidentified speaker (UNKNOWN): a

32:15Unidentified speaker (speaker_0): vulnerability? Because there are systems uh that they were developed uh in a legacy Word world that they wouldn't scale. They wouldn't scale in the world of AI. You can see that recently the security shares are going down, the stock change. So the question is first of all, we need to kind of observe where they are. And then we need to talk to them what good looks

32:45Unidentified speaker (UNKNOWN): like. And

32:45Unidentified speaker (speaker_0): then there are three steps, right? The step number one: if they have enough value. visibility into the new type of vulnerabilities.

32:56Unidentified speaker (UNKNOWN): What

32:56Unidentified speaker (speaker_0): does it mean? Um, PHP happened, and uh

33:01Unidentified speaker (UNKNOWN): the

33:01Unidentified speaker (speaker_0): scanners of today didn't pick it

33:03Unidentified speaker (UNKNOWN): up

33:04Unidentified speaker (speaker_0): as quick as possible because

33:07Unidentified speaker (UNKNOWN): they

33:07Unidentified speaker (speaker_0): need to have signature, the signature was

33:09Unidentified speaker (UNKNOWN): not

33:10Unidentified speaker (speaker_0): uh existing. So do they have like this capability of understanding the new zero days

33:17Unidentified speaker (UNKNOWN): in

33:17Unidentified speaker (speaker_0): their environment? And then next one is which chains, attack chains, are actually they are exposed to. Here we use AI to understand between all the different posture tools which attack chains they are exposed to. The third piece is what we can use today in order to mitigate. the biggest percentage of chunk of vulnerabilities. We found this is a real number, what I told you. We came into the big, big, big, big retail company in the US, Fortune 2 digits, and we found 36 million vulnerabilities.

33:57Unidentified speaker (UNKNOWN): 36

33:58Unidentified speaker (speaker_1): million.

33:58Unidentified speaker (speaker_0): 36 million vulnerabilities. 36 million vulnerabilities, right? Recently af I think you heard

34:04Unidentified speaker (UNKNOWN): uh

34:05Unidentified speaker (speaker_0): the report from Bank of America that they needed to hire 1,000 people in order to fix vulnerabilities that Mythos was finding. 1,000 people. So this is

34:15Unidentified speaker (UNKNOWN): for

34:15Unidentified speaker (speaker_1): that one release.

34:16Unidentified speaker (speaker_0): Just for that one release. So they had thirty-six uh million vulnerabilities. They cannot fix it

34:21Unidentified speaker (UNKNOWN): all.

34:22Unidentified speaker (speaker_0): They cannot fix it all because then there is no business, right? So we need to check now their mitigative controls that they have already in place and match that between the vulnerabilities. This is what we call compensative controls or the concept of mitigation. And in most of the environments, we find that with the existing tools, they can take down up to 90% of vulnerabilities. That means less and less attack chains.

34:52Unidentified speaker (UNKNOWN): Uh

34:52Unidentified speaker (speaker_0): now. Now that 10%, what they can do, right? And here is the risk appetite of the customer.

35:00Unidentified speaker (UNKNOWN): For

35:01Unidentified speaker (speaker_0): the risk appetite, they can we can do something fast and we can do all the human work that they were doing faster and more automated by AI. That means this person communicating to this person over the risk score and then opening it. a ticket to remediate vulnerabilities. Today it's autonomous workflows and it's happening, or we can do a further step. Magic button, right?

35:25Unidentified speaker (UNKNOWN): There

35:25Unidentified speaker (speaker_0): was a vulnerability, collapsed that attack chain. That further step, technology wise, you don't believe it, but we developed it four years ago. Four years ago we had that further step. steps, like we call it agentic remediation. When we talked uh to prospects about that, they were first uh very impressed, and second, no no, we don't, we cannot, we cannot do

35:51Unidentified speaker (UNKNOWN): it.

35:51Unidentified speaker (speaker_0): Like what is it? Like, where's the people? And it's like it's magical.

35:58Unidentified speaker (UNKNOWN): Yes.

35:58Unidentified speaker (speaker_0): And from April's 8th, eighth, Mitos happened, April 8th. and after Hugging Face and after recent uh Bayance hack, we see more and more people using that technology, right? That rujenic remediation. I can tell you even an example of an um energy company, oil and gas company in the West Coast. It was not my bet that this was this is going to be one of the first companies that they go fully agentic remediation. Because they don't have any other chance. They don't have the people, the security forces that a bank has, right? So they connected a lot of their uh environments to agentic remediation. And more than sixty percentage of those attack chains are being remediated autonomously. So I would say first understanding where they are, second understanding if the visibility is right and fit to AI.

36:52Unidentified speaker (UNKNOWN): Uh

36:53Unidentified speaker (speaker_0): third understanding what um attack paths are available, fourth, what can be mitigated by their existing systems in order to reduce uh time to fix, and five using uh agentic workflows flows and agentic agent tech mitigation.

37:11Unidentified speaker (speaker_1): I'd say six is buy really good insurance.

37:15Unidentified speaker (speaker_0): Actually it's interesting what insurance does. We are recently getting a lot of um calls from insurance companies to come and assess a security of a company before actually they do them a premium, right? So it looks like also the insurance companies don't know.

37:36Unidentified speaker (UNKNOWN): I

37:36Unidentified speaker (speaker_1): imagine the reinsurance companies are going to be very interesting. interested in this as well. Because it's it's a big opportunity for them. Because one of the things I'm trying to square in this conversation is if no one's ready for it and if it's too expensive to fix, where does that economic cost, who does that go to? So either the company has to charge more, higher margins, which may or may not the price may may not be elastic. Either they need to bring down their own cost in terms of they need to fire people because they have no other choice, they have to bring that

38:10Unidentified speaker (UNKNOWN): down.

38:11Unidentified speaker (speaker_1): Or s essentially the financial markets has to take over that through insurance, or the US market has to somehow subsidize

38:20Unidentified speaker (UNKNOWN): the

38:20Unidentified speaker (speaker_1): entire the entire economic market, which would be uh especially with the debts.

38:26Unidentified speaker (UNKNOWN): It's amazing.

38:27Unidentified speaker (speaker_0): This is why you are an investor and I'm just an entrepreneur doing innovation. It's very interesting what you said. I think there is a lot of options about like where the money comes from and how we create more with less at the end. But I do believe that at the end, l at least at the beginning of the of this journey, the governments uh should supervise and subsidize stuff.

38:55Unidentified speaker (speaker_1): I think that's why I keep coming back to this deterministic path. You you just mentioned it with the oil and gas company, with you had to go through hugging face and mythos in order for them to take action. I think just human beings, there's such a friction to take dramatic action that there has to be these Chernobyl Chernobyl-level level events that that constantly happen to really get people to take action, unfortunately.

39:19Unidentified speaker (speaker_0): Unfortunately, yes. Uh, I still think in cyber we won't wait

39:24Unidentified speaker (UNKNOWN): uh

39:24Unidentified speaker (speaker_0): for the Chernobyl to happen before it's going to change dramatically, because I think that people

39:31Unidentified speaker (UNKNOWN): uh

39:31Unidentified speaker (speaker_0): that are developing the technology, but also people that are on the defense side,

39:38Unidentified speaker (UNKNOWN): they understand.

39:39Unidentified speaker (speaker_0): understand that um when the Chernobyl will happen in cyber, it will be too

39:45Unidentified speaker (UNKNOWN): late

39:45Unidentified speaker (speaker_0): to fix it. Like

39:47Unidentified speaker (UNKNOWN): I

39:48Unidentified speaker (speaker_0): don't know how to tell you this, but there is no way you can monitor something after it will it's already widely developed and in place in i everywhere. Right? You have to do it now.

40:01Unidentified speaker (UNKNOWN): I

40:01Unidentified speaker (speaker_1): went to dinner a with Morgan Stanley's chief security officer and he said, The reason you're talking to me is because there hasn't been a vulnerability yet. If there is a vulnerability, and he had been there for like seven, eight years, I believe, and it he's like, If if there's ever a problem, I won't be at the next dinner.

40:18Unidentified speaker (speaker_0): Yes.

40:18Unidentified speaker (speaker_1): And it it is this kind of existential thing. It's it's it's a difficult career because essentially you have your job until you don't have it.

40:25Unidentified speaker (speaker_0): David's

40:25Unidentified speaker (UNKNOWN): very stressful

40:26Unidentified speaker (speaker_0): every day. It's every day. I can I can give you that. Maybe you add it after that to uh to the podcast. Every day we have a zero day. There was days that zero days was like once a week, zero days was once a year. Every day there is a big thing with a new zero day. Zero day is those vulnerabilities that they don't have yet

40:46Unidentified speaker (UNKNOWN): uh

40:47Unidentified speaker (speaker_0): effects a fix. this is what zero day is, right? And uh um every day, every day. Like being uh a security leader, it's it's very hard now. And it's uh

41:01Unidentified speaker (UNKNOWN): we

41:01Unidentified speaker (speaker_0): should thank these guys or gears uh that they are still doing it.

41:04Unidentified speaker (speaker_1): Speaking of these guys and girls, these chief information officers at these companies,

41:10Unidentified speaker (UNKNOWN): in

41:11Unidentified speaker (speaker_1): many ways they're probably the most sympathetic to these issues and probably the least controversial customer of these problems. You talk to them every single day. What are they saying?

41:23Unidentified speaker (speaker_0): There was uh a meeting we had with 30 different uh chief uh information officers and chief information security officers, CISOs and CIOs. This is how we call them. And there was one moment of being completely transparent. And uh actually it was uh a lady CISO that said, I wanna say that for the first time, I don't know what bad looks like. Therefore, I have no idea what I have to do to be good against it. Like for the first time in the history and she's like a CISO So 30 thirty years, right? And this is uh exactly what AI did. to this world. So first of all, I think what it makes them doing, suddenly there is much more coalitions and and partnerships between the different entities. And I very, very welcome this, right? I think one of the biggest challenges in the US

42:24Unidentified speaker (UNKNOWN): that

42:25Unidentified speaker (speaker_0): you don't talk to each other that much.

42:28Unidentified speaker (UNKNOWN): For

42:28Unidentified speaker (speaker_0): example, in Israel, it's a very small country, and they everybody know each other from the military service or the from the reserve, and everybody talk to each other, even stuff that they shouldn't. And here they don't talk, and I see like at least four different companies being attacked by the same vector

42:47Unidentified speaker (UNKNOWN): at

42:47Unidentified speaker (speaker_0): the same time. I'm like, why don't you you talk to each other? You're from the same sector. Like why? Just take the phone and

42:55Unidentified speaker (UNKNOWN): or

42:55Unidentified speaker (speaker_0): signal and just call this guy. So uh first of all I think there is much more collaborations and coalitions actually. Uh some of them we are also involved um with the Frontier AI uh labs but also without so this is one.

43:13Unidentified speaker (UNKNOWN): This

43:14Unidentified speaker (speaker_1): is a cultural difference.

43:15Unidentified speaker (speaker_0): This is very cultural difference.

43:17Unidentified speaker (UNKNOWN): I've

43:17Unidentified speaker (speaker_1): noticed that as well. I'm immigrant as well. And I always meet with people and I'll say, Hey, here's the twenty percent of the market

43:26Unidentified speaker (UNKNOWN): that

43:27Unidentified speaker (speaker_1): we're competitive in. Zero sum. Let's not talk about this. Let's talk about this eighty percent white space where we could collaborate or potentially be helpful. I think people people are always surprised by the directness. of that.

43:39Unidentified speaker (speaker_0): I see much more startups. They are not anymore relying just on the legacy tools that they were used to. They understand that they need like innovation and uh this innovation not going to happen by thirty old comp t year old companies that fast. So even uh companies uh that are bankings and have regulations, and before that they were very like cautious with letting a startup to come

44:11Unidentified speaker (UNKNOWN): in.

44:12Unidentified speaker (speaker_0): They work with a lot of different startups, and I like it a lot. So there is much place for innovation. And the third I see, which is interesting, is the CISO is also telling me this. Suddenly there were their workforce is becoming on board much faster than before. You know, the biggest problem of security in the US, especially, but not just in the US, is the skill. You don't have this post-NSA people that are free enough now to work in a bank or in a retail, right? You don't have that workforce. So what to do?

44:50Unidentified speaker (UNKNOWN): So

44:51Unidentified speaker (speaker_0): suddenly you see junior people in that they do the first steps in security, use agents and they are become like cybersecurity defenders, like beast mode. So I like it very much uh to see these people that have a chance to uh to get much more value and much more compensation at the same time.

45:12Unidentified speaker (speaker_1): You're at this famous part of military 8200 in in Israel. You were there for quite a while, you became a one-star commander. What do you take from that experience to what you do today?

45:27Unidentified speaker (speaker_0): That you should never get comfortable with the place you

45:30Unidentified speaker (UNKNOWN): are.

45:33Unidentified speaker (speaker_0): get comfortable it's too late. I think one of the

45:38Unidentified speaker (UNKNOWN): big

45:38Unidentified speaker (speaker_0): basics of what we learned uh in the unit and the people that are actually being recruited are people that they are not happy ever with the results and they always like for look for other ways to make it better, to make it faster, uh with or without AI. So this is uh definitely a lesson learned uh from there. And you should always, I think, hire people that they are better than you and they are much more different than you. Most of the departments that I led uh in my past, they were the first time departments. Like I was an entrepreneur also. uh there and I always brought like people from the different disciplines and technology and intelligence and others to create this uh task forces that they tackle a very very hard problem. So

46:41Unidentified speaker (UNKNOWN): uh

46:42Unidentified speaker (speaker_0): this I took with me uh for sure. The second piece is mission is above everything. I don't think I could do something that it's not cybersecurity because there is a mission in cybersecurity and there is bad people

46:56Unidentified speaker (UNKNOWN): on

46:57Unidentified speaker (speaker_0): the other side and I have to win. Like I couldn't work 24 hours, including a Saturdays, Sundays all the time, without understanding that there is an evil that I have to win. Uh it's not just about the economy and the money. It's about like doing the right thing at the for the right people. at the right time. And I think this is also a value

47:19Unidentified speaker (UNKNOWN): uh

47:19Unidentified speaker (speaker_0): that I took uh with myself

47:21Unidentified speaker (UNKNOWN): from

47:22Unidentified speaker (speaker_1): this anti fragility.

47:23Unidentified speaker (speaker_0): Yeah, true.

47:25Unidentified speaker (speaker_1): It's the paradox of Israel and Israeli people is the more people attack them, the more anti fragile they become.

47:31Unidentified speaker (speaker_0): I can definitely

47:33Unidentified speaker (UNKNOWN): uh

47:33Unidentified speaker (speaker_0): talk about this. Uh still there were moments also in my past uh especially after I I became a mom, not before that, that I could feel fragile, even uh in the army. And uh there there's there are still moments uh in the world security that uh I feel

47:55Unidentified speaker (UNKNOWN): it

47:56Unidentified speaker (speaker_0): on my skin even I'm not there anymore. Um so definitely it's something that you take

48:03Unidentified speaker (UNKNOWN): uh

48:04Unidentified speaker (speaker_0): with yourself uh forward

48:06Unidentified speaker (UNKNOWN): uh

48:07Unidentified speaker (speaker_0): to your home, to your work

48:09Unidentified speaker (UNKNOWN): and

48:09Unidentified speaker (speaker_0): to the who you are.

48:10Unidentified speaker (speaker_1): If you go back to where when you had just co-founded Zapron just several years ago, probably decades in AI time, and you could give yourself one timeless piece of advice, what would that be?

48:21Unidentified speaker (speaker_0): Don't listen to everybody everything that the investors tell tells you. When you're doing your first steps uh in entrepreneurship, when it's the second or the third time you already know. But for the first steps you're a little bit lost and you're looking for guidance. And the first like VC that invest in you, it becomes almost your God, right? So and I think um there is something so powerful not knowing the rules and being like almost very stubborn about what you believe. And I think this makes good companies. And I have examples of that at the very beginning of the time. So

49:03Unidentified speaker (UNKNOWN): uh

49:04Unidentified speaker (speaker_0): I think it's a good spice to have for the new

49:08Unidentified speaker (UNKNOWN): uh

49:09Unidentified speaker (speaker_0): ones coming. Also, since technology is not the big barrier here, right? Like

49:15Unidentified speaker (UNKNOWN): um we

49:17Unidentified speaker (speaker_0): have our uh office dog in Israel, his name is Mosh Mushu. And we always always laugh. Uh I don't know the type, but he's fluffy and and beautiful. And uh we always tell that uh he's probably have his own startups, like he is working with a couple of agents and make starts startups. up. So the problem is not tech, the problem is those people that they know what they want to achieve and they are flexible and agile, but not fragile, as you said, and they wanna move like with the technology and almost surf over the

49:58Unidentified speaker (UNKNOWN): um

49:59Unidentified speaker (speaker_0): waves of the technology to achieve the value at the end. Like being super, super, super focused on the value because the world is becoming so much noisy that um also knowing what not to do and which are the values uh and the impact you want to make uh for which crowd is very important because the moment you defocus your yourself, it's going to be lost. So listen less to investors and be agile

50:36Unidentified speaker (UNKNOWN): and

50:36Unidentified speaker (speaker_0): mission oriented. Well,

50:38Unidentified speaker (speaker_1): it's been an absolute masterclass. Thanks so much for stopping by.

50:40Unidentified speaker (speaker_0): Thank you so much. I enjoyed it a lot.

This page was produced automatically: the audio was transcribed with nvidia/parakeet-tdt-0.6b-v3+sortformer-v2.1 and interpreted by Qwen3.6-35B-A3B-UD-IQ3_S.gguf. Transcription and interpretation can contain errors; always confirm against the original episode. Nothing here is investment advice.
All insights