The AppSec AI Revolution – Apiiro x OWASP — Transcript

Panel is the apps AI revolution, and on the one hand, it seems like a very stereotypical concept, a very sort of, oh, you know, AI is revolutionizing, AI is the future, whatever else. But certainly, as someone who has been involved in software and software security over the last year, it has been a wild, wild year. I mean, it's hard to believe that Claude Code, for example, was released less than a year ago when it seems to be absolutely everywhere right now.

software and software security over the last year it has been a wild wild year I mean it's hard to believe that claude code for example was released at less than a year ago when it seems to be absolutely everywhere right now.

Everyone seems to be using it. Everyone seems to be adding add-ons to it, and, you know, less than a year old. We've seen things changing week by week, month by month. New models being released, new mechanisms to use it, people spinning up brand new things just by letting the code generator run over and over and over. And, you know, certainly from a software security perspective, it raises a lot of questions. Certainly questions I've asked myself over the last year, and I've tried to sort of bring some of those questions together for our panel today. So before we get into that, I would ask our panel members to maybe briefly introduce themselves just in case anyone's not familiar, and then we'll get into the questions. So, ladies first. Go for it. Um, take, you have it on. Yeah, you can hear me.

spinning up brand new things just by letting the code generator run over and over and over. And you know certainly from a software security perspective, it raises a lot of questions. um certainly questions I've asked myself over the

Yes. Okay. Hi everyone. I'm Kenazi. Today, part the managing partner at Kahin Venture Capital, the Israeli CISO's VC, and a former SVP security engineering and CISO of Jrog, and prior to that, Plica and product manager for 10 plus years before that. So glad to be here. Thank you.

familiar and then we'll we'll get into the questions. So um ladies first. Go for it. Um take you have it on. Yeah, you can hear me.

Good evening. Thank you all very much for being here tonight. I want to give a shout out to team OASP and thank you again to all of the Aamites in the room that stayed tonight to be a part of this, as well as our partners from Piro. I do also want to say very much a thank you to everybody in the room for your commitment to stay in the fight and to work these complex problems. What you do every day matters. It matters now more than ever because the threats and the speed of those threats are getting faster and faster. And tonight's topics are really important in regards to how we're going to show up and how we're going to address things. My name is Shawn Martens. I am super humbled and proud to lead an amazing group of lawyers in our infrastructure and application security business.

years before that. So glad to be here. Thank you. Good evening. Thank you all very much for being here tonight. Uh want to give a shout out to team OASP and thank you again to all of the Aamites in the room

Wow. Thank God I lost before. [Laughter] That's it. Hi everybody. My name is Thomas. I'm the former CEO of GitHub until last October. Last August, I was at the helm at GitHub for about four years. I'm originally from Germany, as you can hear. Usually, people figure that out when they just say hello in the skill. I don't know how. And I work with a, that's why I'm here this week. I'm also working on a new startup, and I love to say I'm just a software developer, and I love coding, and I never loved it more than now.

work these complex problems. What you do every day matters. It matters now more than ever because the threats and the speed of those threats are getting faster and faster. And tonight's topics are really important in regards to how

And you invented GitHub Copilot, which is number one in the AI coding agent.

Wow. Thank God I lost before. [laughter] That's it. Hi everybody. Uh my name is Thomas. Um I'm the former CEO of GitHub until last October. Uh last August I was at the helm at GitHub for um about 4 years. Um

No, you said no. Then you said I'm responsible for many of the things we discussed tonight. So, okay.

here this week. Um I'm also working on a new startup and I love to say I'm just a software developer and I love coding and I never loved it more than uh now.

Thanks much, folks. So the first question I wanted to talk about was, I guess, around application security risk. Security is a big field. You know, as a CISO, there is a lot to worry about, there's a lot to think about. Application security is, you know, we're here to talk about one aspect of that. So, I guess, the question I wanted to raise now is, you know, where does application security rank compared to all the other risks that maybe are keeping a CISO up at night? And has that changed over the last year given what we've seen with both AI in use in development and AI in products themselves? So, you know, where does it rank and how has that changed?

No, you said no. Then you said I'm responsible for many of the things we discussed tonight. So okay. Um thanks much folks. Um so the first question I wanted to talk about was I guess around application security risk. Security is a big field. you know

All right, I'll start.

wanted to raise now is you know where does application security rank compared to you know all the other risks that maybe are you know keeping a CISO up at night um and has that changed you know over the last year given what we've seen

Yeah, sure. So, I can talk about the complexity of being a CISO in the era of AI developing. You know, everything runs so fast, and developing becomes something that every engineer, every software engineer is now a team lead of agents, right? It's not practically writing, almost not writing code, but running agents that develop code. So everything becomes so fast, and as security gatekeepers, you truly, truly rank the application security and the supply chain in general as almost one of the three major risks in cybersecurity because supply chain attack can be something that evolves companies and nations, and that's become something super, super urgent for all of us. So it's evolving, and I can share that we spend a lot of time and effort.

Yeah, sure. So, I can uh talk about the complexity of being a seesaw in the era of AI developing. You know, everything runs so fast and developing becomes something that like every engineer, every software engineer is now a team

Thank you. I can share that we spend a lot of time and effort in order to make sure that the supply chain security will be so strong with gatekeepers, not blockers. So part of that was adopting the best solutions out there in the market and making sure that every vulnerability will be fixed.

as almost one of the three major risks in in cyber security because supply chain attack can be something that evolve companies and nations and that's become something super super urgent for all of us. So it's evolving and I I can

Is that something that's changed over the last year?

gatekeepers not blockers. So part of that was adopt the best solutions out there in a market and make sure that every vulnerability will be fixed. Is that something that's changed over the last year?

Yes, definitely, because it's been accelerated. One of the KPIs that we thought of is just block releases with critical vulnerabilities. That was a KR. That was something that we would truly try to block. You cannot do that anymore. You just need to narrow it down too much. Something that came, and that was a huge challenge. And we saw that it's become almost impossible. So we truly invest in startups and a lot of self-development just to make sure that we'll be able to keep the pace. I think maybe before even talking about risks, I think that when an agent writes the code now and introduces new APIs, new dependencies, new data models, new technologies, and expands the attack surface, we don't have the fundamental thing, which is inventory, to know what these agents introduce. I'm saying you are developing so fast, and you do not read or review. You have like a yellow mode. No one reviews the pull request, and they want to approve it and move fast to production. So even before talking about vulnerabilities, understanding the attack surface and the amount of resources, code resources that are getting into your code is a huge problem. If you compare it to cloud security, this was the first thing that they've done: building an inventory of everything that you have in the cloud, understanding the cloud architecture. No one is doing it in application security. And this is the challenge. The second thing is that when the velocity is growing four times from the last research that we did with Fortune 20, the amount of vulnerabilities are growing so fast that you cannot ask your developers. It breaks the physics of apps. You cannot ask the developers to deal with more and more fixing vulnerabilities in every sprint. So we need a fundamental shift from detection prioritization to preventing it.

that anymore. You just need to narrow it down too much. Something that is is is came and that was a huge challenge. And we we saw that that it's become almost impossible. So you we truly invest in startups and a lot of self-development

Agree. Do you want all of us to answer? We'll be here all night. [Laughter] I mean, Akim is obviously a large organization.

expand the attack surface, we don't have the fundamental thing which is inventory to know what these agent introduces. I'm s I'm saying you are developing so fast and you do not read or review you have like a yellow mode no one reviews the the pull

Akami's got a lot of things to worry about at the same time. But this is something that has now suddenly become quite a huge and emerging threat. So it'd be interesting to know if things have changed over the last year and how you're seeing that manifest within the organization, within the organization's priorities, within the organization's budgeting, within the way the organization relates to application security. So I would say absolutely. I'll use an American term, which is we are living in the wild, wild west. So what has changed is the fact that we have a rise, just like we all experience with shadow IT, to the point that Adon made number one what is the visibility of the actual use of these agents within the organization, and then number two, how are you going to identify drift? And more importantly, where we need to get to is how are you going to take proactive actions to stop that. So that's the internal aspect. What our customers are also worried about is it takes 42 seconds. 42 seconds to successfully launch an LLM fro...

security, this was the first thing that they've done building an inventory of everything that you have in the cloud. Understand the cloud architecture. No one is doing it in application security. And this is the challenge. The second thing is that

when the velocity is growing 4x um from the last research that we did with Fortune 20, the amount of vulnerabilities are growing so fast that that you cannot ask your developers it like it's breaks the physics of apps.

You cannot ask the developers to deal with more and more fixing vulnerabilities in in their every sprint. So um we we need a fundamental shift from detection prioritization to preventing it.

Agree. Do you want all of us to answer? We'll be here all night. I [laughter] mean Akim is obviously a large organization.

Akami's got a lot of things to worry about at the same time. But this is you this is something this has now suddenly become a quite you huge and emerging threat. So it' be interesting to know you if things have changed over

the last year and how you're seeing that manifest you within the organization within the organization's priorities within the organization's budgeting um within the way the organization relates to application security. So I would say uh absolutely um I'll use a an American

term um which is we are living in the wild wild west. So what has changed is the fact that we have arise just like we all experience with shadow IT to the point that Adon made number one what is

the visibility of the actual use of these agents within the organization and then number two how are you going to identify drift and more importantly where we need to get to is how are you going to take proactive actions to stop

that. So that's the internal aspect. What our customers are also worried about is it takes 42 seconds. 42 seconds to successfully launch an LLM from a DOS perspective. According to industry analysts, breaches now go hidden for 80 days based

on some of the tooling that's available. So think about that. There could be things that are growing and excfiltrating data within your environment that you thought was a known good, which is a known bad, and now it's taking you 80 days to uncover that. So,

I believe that discovery and inventory is absolutely critical, but observing what's happening and identifying that drift and then proactively as quick as we can to mitigate those risks is important. Why? SLAs's used to be okay at 15 minutes, 10 minutes. Now SLAs are

the speed of a prompt milliseconds. I think the counterpoint is you know everybody talks about the rising risk in AI writing all this code. It also has never been easier to find out about OASP. Uh it's never been easier to

actually put the over 10 I think I first discovered that like 20 years ago, 15 years ago or something like that to actually apply that to your source code, right? You can just ask your favorite agent uh in the sidebar uh to uh figure

out why this code has a SQL injection. I think for me one of the eye opening moments was um a few years ago maybe two or three years ago um not only could it um find the SQL injection it could

actually explain the SQL injection to you or you know your co-workers and then you could even ask it to show me how an attacker would exploit the SQL injection and you could never do that before without AI right you could read all the

documentation you could see some sample code but it wouldn't show you the relevance of that for your source code and of course then you can also say fix it for me. I'm pretty sure actually you can probably just tell plot code to look

for over top 10 findings in your source code and then try to fix them or or show you how to fix them. So I think both of these things are true. It's always the cat and mouse game. Of course all the

attackers have AI available but we also have these AI tools available to us and we just need to use them to make our source code more uh secure. Mhm.

I would add to that not even a step before there are probably architect security application security architecture. We now AI narrowed the time from idea um to developing so fast. So think about threat modeling that we used to do for

major features or risky risky features that really related to authentication and authorization things that we used to review and think about them and negotiate and talk about them you don't have time to do that tomorrow morning it will be developed already so you need to

move very fast and I totally agree with it about a fundamental change we need to be there with agents that will accelerate our ability to reviewed as well.

Yeah, I think we're certainly going to explore that a second in the in the next question. Um but I guess and and and you think next question we're going to talk sort of more about you how we basically

use AI to try and accelerate what we're doing um potentially building things in house and and you before we get to that I guess that's you are we seeing um are we seeing a situation where maybe in previous years um people were trying to

sort of get budget for apps investment or absc headcount to actually sort of sit and do sophisticated things um but they were struggling Are we seeing that you know having just had the past year of massive massive acceleration in code generation whatever

else are we seeing that that budget is now more forthcoming are we seeing that appsec is now sort of being pushed higher up in the priorities you know the business stuff you know I think there's always this discussion within apps oh

you constantly fighting against all the other priorities okay we need to get new features out or we need to comply with this regulation or you know whatever it is that we need to do are we seeing that now suddenly sort of the actual concept

of checking for vulnerabilities and avoiding building vulnerability in the first place. Is that now getting um is it now easier to pull budget out for that sort of thing? Do you think that's what they're saying?

You're holding the budget now. Go ahead. I I would just say that uh it it is a compelling event. Um it changed the priorities of the again I'm talking from my point of view of working with CISOs in [clears throat] Fortune 2000

companies only. Um and they literally in the last year shifted budgets towards application application security um and we see that it it becomes you know ups I'm 24 years in cyber um specifically obssec um it always was a topdown

process I need to find out h the relevant security engineer to convince him that we have the new shiny tool to solve a different problem and it was very very hard to get to the sea level to get the budgets now it's a top- down

approach where the CTO CIO and CISO are going down to the team and saying hey what how we are dealing with the expansion of the attack surface how we are dealing with reviewing all this, how do we prevent these risks from reaching

production and we saw an inc tremendous we said it uh publicly that we grew over 100% because of this demand and we don't need to convince anyone that there is a problem which was the case in ASBM you

know um so I think the evolution of apps was a like scanning detecting vulnerabilities manage the risk and now it will be um preventing it.

Okay. I think if you look at you asked about headcount but you could also just you know look at budgets all up um which are basically abstract s it's an opx operational expenses as opposed to cost of cost sold like running a service and

uh most of the you know tech companies certainly in the United States have not seen increases in their budgets. Uh in fact most budgets have stayed flat or slight slightly increased with you know cost of uh uh salary increases bonuses

and whatnot. And um within those budgets we certainly see a shift um uh in thinking um if you if you think about you know the cost of your employees or of your own cost at your company obviously there's your salary and your

bonus and you know depending on what country you're living in benefits and taxes and all all that um what what has and then then there tools in your office and and maybe have a rental car or so but what has been rising

massively over the last year is your token cost right you cost if you're a researcher right and and that is very interesting because there's a direct function of your productivity tied to how much you're using an agent right and

and and your budget as an engineing manager typically is flat over the year or you're fighting for like a small increase um or you're fighting with somebody in your organization to get something from them, but [clears throat] it certainly isn't tied to how much your

developers just shipping, right? And so you might actually get into a place nowadays that the more productive your developers are, the more token budget you need, but you don't have that because the budget is flat, right? And so we're seeing this almost like

inversion similar to not inversion but like similar behavior as with our service. The more successful our service is, the higher our fs are and hopefully our revenue also goes and then the CFO is fine with that of course going. But

on the operational side that typically wasn't the case. More shipping didn't mean the CFO gave you more budget. If anything, they said well you got along with the budget you have so find more optimization. So I think overall we we

don't see massive shift on the total on the top line of these budgets but we see a shift in allocation and we need to start thinking about okay so how do I allocate my token budget like in my

startup I think they're spending between sometimes between $50 and $100 per engineer per day on on a on a topic tokens right if you do the math 200 something work days a year that's $20,000 per engine year per engineer per

year and that's only going to go up right and so We got to find ways to optimize that in our planning processes.

And then also, you know, and you're alluding to this is how much of that is used for code review, abstract review, quality review, all those kind of things was just generating more and more code and not being able to review all that

and basically that becoming the bottleneck. Mhm. Yeah. [clears throat] And I think that's that's a very nice segue into our next question, which is um u around sort of internal apps teams.

How are you seeing internal apps tech teams take advantage of you know AI assistance to actually build internal solutions? I think that um you know maybe once upon a time you know if you could you want to do apps and sit like

writing code manually to solve a particular problem you that was a way of solving a solution solving a problem but that was quite labor intensive and often it would be easiest to try and bring something off the shelf nowadays with AI

code generation suddenly doing that is a a lot faster you know I've done this myself multiple times over the last few months precisely because it's now so much easier so I guess the question is how you know how you what sort of

solutions or What situations have you seen internal apps teams taking advantage of of this to build solutions to internal apps problems?

I I can I can share what's how I see it. Um basically today it's AI or die. Okay.

So you cannot think about should I adopt it as an ABSAC. You have to do that. You have to you don't have other choices. So that's the reality. Um the fact is that we truly appreciate great product, great

solutions and prefer to buy versus develop. We are not developing because it's cool and it's easy. Um just because we need to give context. We need to enrich the solutions that exist with our own context to make things much more

efficient. uh for example for that maybe for the app six team the teams and the penetration testers here is around find the owners like who I found a vulnerability but this I know a little bit about the service um in the API but

who is the owner where should I go to and and reach out the person that will fix it as as fast as we can especially when teams are changing so much so fast uh in in agile so thinking about use of

AI the advantage of that this is urgent it's something that we have to do um myself I I cannot someone will take like the agent out of me today I don't even I cannot even think of writing emails

anymore or answering or writing my bio for for that meetup or anything else thinking about PR that I need to release without AI without a double check but It's it's not replacing but it's just extending my ability to to do much more

things much much faster. We need to invest in that put our knowledge and train the agents to do that fast. So this is that's my take AI or die.

What I would say is that um [clears throat] there were a bunch of non-believers 9 months ago, 12 months ago in regards to what the feature set and functionality is. Everybody's a believer now. So because everybody's a believer

now, there is this ground swell of how can I leverage this tool to actually make things faster, to be more creative, to get to the end result. And to Thomas's point, I believe what we're seeing and going to continue to see is,

okay, you're running your business. Here's your budget. I could go hire 10 interns and assign those 10 interns to my lead development teams or my engineering teams. Or I could spend $20,000 a year and I could leverage Claude and I could use a very

sophisticated person who has an understanding can really drive that Claude instance for higher production and outcomes. So I think that the paradigm shift is this is going to continue to accelerate each generation and get better and better. And obviously

we have to pay attention to the the costing models as they shift. But I do believe this will be augmentation of very sophisticated drivers until we get to pagent. You just made an implicit um uh um statement that interns are going

to replace with claude. I actually think that's the wrong assumption um because what we are going to actually see is the rise of the junior developer because similar to um different uh technology changes in the past what's going to

happen is that generation of young folks that are growing up now you know I have two boys 131 and they're using more AI than probably the average person uh in in Germany or the United States and um we're going to quickly see students

coming out of universities that are so used to uh you know um leveraging the power of AI that they are going to not go through the same culture change that you all have to go through of adjusting your workflows that you're used to for

the last 10 15 20 years um and as such the interns actually as always as they have done in the past bring this new energy you know the new knowledge new how do I use AI for abstract back into

the company so you're absolutely right we're going to uh see scenarios where we are replacing kind of menial work with the help of AI. But those folks, those the young folks in the same way that they are so much more knowledgeable on

um you know filming a Tik Tok of their vacation. Well, I find it awful to [laughter] like this is like the next generation always grows up the technology that we find unfamiliar and as such they're so much more familiar with it and so much

more powerful with it that we will have to embrace that uh and bring those young folks into the company and say you're no longer a junior developer from 10 years ago. you know at the skill set of a

senior developer and as such show us the way we go back to the original question for a second um and I would I would say I would just say that we we don't you ask if apps teams are actually using AI

to develop the tools we we don't see that across these large enterprises because of exactly what you said there is a policy that you cannot develop any tool that you want. Okay. Um and to develop a tool in a large enterprise, you need actually

budget process go to Yeah. And but but what you said goes back to the fundamentals and inventory. You want to understand who is the owner of an API or open source dependency or a code module or microser or Yeah. and I added a new genai

framework or I developed an AI agent or whatever. Um and and this is this is I think what the appsec today are doing with AI trying to understand who are the owners, what is the software architecture.

All the the developers here that are actually using AI coding agents, they do not understand not because they are not professionals because the velocity of the changes doesn't allow you to understand the software architecture and it changes so fast. So you need

something another agent to understand that. Um and from our experience, this is what we see abs up abset engineers are doing with AI trying when they go get into a threat model or into a security review or helping fix a

vulnerability they try to understand how the software architecture looks like instead of flying blind. Thank you. And he invented it today. uh but flying blind I like we really feel that if you don't understand the software architecture how can you pro how you can

give an advice to a a developer that had a vulnerability without understanding and looking at the bigger picture um and I think this is a must to be good at our job as ABSC engineers yeah I mean certain yeah certainly the

situal situational awareness thing is is a big deal especially you know large organizations it's a lot harder to figure out okay who owns this who does this belong to? I think you know maybe the it's possible you know the the the

sort of DIY tools or DIY code maybe more in in smaller organizations where they are less organized. That's certainly been my experience in some of the some of the you less enterprisey more sort of like startup plus organizations that

where that's where they start doing things um slightly differently and yeah I'm going to use this as a segue into the next thing because on the other hand you know situational awareness works in a few different ways and you

know in principle large enterprises have policies saying you can only do this you can't do this you can you know you have to do this in this certain way Um but on the other hand I think you know

again AI and LMS have very much sort of democratized what you can actually do how you you know can you actually do something yourself um you know there's less you know and let's scope this within sort of traditional engineering

teams for now or traditional let's say teams for now but obviously you know suddenly there's a lot more possibilities you know people can sign up for uh you know I don't know you know their own personal claw subscriptions or

they can um you know download an LLM off the internet and suddenly integrate it as a new product feature. So, okay, well, now we're not getting built for it cuz they downloaded the whole LLM and now they're just using it in the

product, but it's still you've now got product feature that's based on this LLM. Um, so I guess the question becomes, you know, so what visibility and dare I say control? I mean, we can definitely have visibility control is

obviously a little bit tricky. You know, what what mechanisms are you seeing um to sort of deal with this sudden rise?

And you know I guess you how if what sort of mechanisms are you seeing? How effective are they or are you see are you seeing situations where something new needs to be brought in to try and get control of this? I think one way to

look at this is that um software engineering is so far ahead on AI because about a decade or so ago we invented DevOps and made our lives harder by introducing all these things that we don't want to do when we write

code when we write code 20 years ago. Right? When I learned coding, I didn't write any unit tests as to to my memory.

There wasn't even such a thing on a Commodore 64, right? Like raise your hand if your first project uh had any testing, any main languages, L detection, CI/CD, none of that stuff was the point of coding your first computer

game, your first calculator or anything like that, right? But we introduced in software engineering all these processes at zero trust you know branch protection or security scanning secret scanning key walls all of these things to prepare us for AI right because we didn't we don't

actually trust our human developers either we know they're making mistakes that way you have to ask somebody in your team to review your code and proof or or maybe find some stuff and we were discussing earlier if you actually think

about how code review works in your own teams it's a it's a a reputation based system like the principal engineer that you've been working with or for the last 10 years or that sits next to you get a

much more linear code review where like ah you know Elan bought all this looks cool looks good to me merge it while the junior engineer that just joined the team gets like the nitpicking on every single line of code right that there's a

space missing here and a semicolon there and so on right and that has nothing to do with actually the risk of introducing a bug it has all to do with you trust one of these engineers more than the

others Right? And AI levels that playing field because we have to introduce the same code review processes, the same secret scanning processes, code scanning, you know, a and whatever acronym you have, you we have to apply all of these to to AI generated code and

if anything, you know, if we your CI/CD run fails, but let's not merge that code whether it comes from a human or or from from AI. And I think that is what makes AI so successful in software engineering. if you have actually all

these things in place. Now there's lots of organizations that don't um you know um whenever um I've been on the world with GitHub and I met with banks and finance in institutions and local governments they're running coal on

mainframes and leaving aside that co is 50y old language from the Eisenhower era. Um 1958 or something like that. Um guess what? Cobra also has no unit tests and it has no CI/CD and it often is not on Git on GitHub and all these kind of

things, right? And so the the problem that we have as an industry are living on this insane spectrum of technology that is out there. co and main things and Windows 95 terminals and PHP and Pearl and all the stuff that was written

in the '90s when the internet was hot all the way to now Python and Go and Lust in the modern languages and almost no organization out there that is more than a year old doesn't have legacy code right like my I think my favorite joke

as a developer is the happiest day is when you start a new project and from there it's all downhill you know able to manage like oh they have all this tech all these new framework works is all We need to refra and it takes us 6

months, right? And and that's the life of a software developer. And so we got to realize that no matter whether the code was generated by AI by humans, we constantly live in the struggle of the entropy that our software projects

refactoring it, cleaning things up, you know, paying down the tech debt to just have new tech debt from the from the latest framework or now from the latest agent.

By the way, uh we are in the industry, we confuse between the risks that are coming from using AI coding agents to develop code versus the risks that are that we introduced by adding Genai frameworks or agent inside our code to

serve our customers. So these are two different attack surfaces. Having said that, they are intertwined, okay?

Because the agent can add an AI or genai framework um and call like an exit point to open AAI and and send PII data from from my software architecture. So I'm just saying we need to differentiate between these two attack surfaces. Um

and there are different controls mechanisms for using AI coding assistant versus controls of using AI in my code.

Um today we see more and more and more um I would say governance early in the source control manager to say you can use only this framework from this version because this is vetted in my organization. So or off for models for a

ji frameworks that calls a third party service or even developing AI agents with specific frameworks. So you have a policy on what you can add into the organization. Why? Because when you try to sell software to a customer, they ask

you for a risk AI risk assessment. So if you go for example to trust.appro.com apo.com publicly available. You see that we have a risk assessment for AI because we cannot sell a deal above $1 million without having a risk assessment and

showing that we have an inventory of all AI in our code and of course that we govern and put guard rails.

Amazing that you said that it's exactly the point that I wanted maybe to share a little practical how we did it. uh um I called this program trusted apps um and I sailed that internally in the organization that wasn't easy but thanks

to the observability tools that we had we willing to explain what we are looking for and how to adopt the gateway the gatekeeper. So we set up uh a policy based on all the things all the top things that we want to achieve from

trusted up and we make sure we started slow and create that for example zero critical for every new releases mandatory to code review to code owners and and much more of course and a part of the release we we managed to achieve

a point that a goal And if we didn't we rank it but gold, silver and bronze two applications and make an internal competitors between them. So gold application will be able to run fast and move fast with at the stations and

that's the most important thing that I wanted to share. Um software cannot be released to production without the stations. Okay. So that won't be uh a golden image that won't be application that is certified to run on production

unless you are bringing at the station that you don't have uh vulner critical vulnerability in place under SLA all the vulnerabil the rest of their vulnerabilities that aren't there SLA must be remediated on time you don't have tokens hardcoded tokens out there

you don't have uh you have to get at the stations of SAS with clean environment uh and of course code ownersh and more and more policies as long as but we build a policy of blocking up to that point that

would was very mature. Okay. So that mechanism allow us to show and demonstrate what what are the gold standard that who are the stars internally and that help us a lot to bring that to the other teams and and

increase the level of the maturity. If I'm not the moderator, can I ask a question?

So um now um we published it uh publicly I think a few months ago we saw in um an organization with 7,000 developers 62,000 uh repos um 10x more vulnerabilities even after you prioritize them with the risk based approach it's

8x okay How did this actually this model that you're talking about? It it actually breaks when when you have 10x more findings and you need to ask the developers to find to to fix them, but then you you need like the business

needs to grow and you need to release more features faster and faster. So the the physics doesn't work if you have 10x more findings on every release. So this is what we see that the old AB apps um

model of the gating based on vulnerabilities or based on generic risk scoring prioritization CVS score EPSS whatever threat it still doesn't shrink the or or minimize the number of vulnerabilities that I I as a developer needs to fix and with AI coding agent is

just growing and growing and growing Because the agent can write code when we sleep, okay? Or whatever.

So just to answer you, the gatekeeper is the end gate to make sure that you have trusted releases, but it's not the solution. It's like runtime versus ship left, right? That's the same. So we chose the right product that are very

close to the source code and find mint fix. We find the stat immediately run to the that and that's goes back to find the owner and report immediately and run PR automatically help the developer to to move fast and yeah we we customize I

I wish that there were amazing solution out there that are really close to the shift lift with no noise at all but we needed to to customize things. Yeah, we did. So the target is to do not find it

over the runtime. Of course not. We were at the very mature level to block but that's not the solution. That's the end of the goal to make sure that we have trusted and it's back to the customers.

I won't release untrusted software at all. I won't do that. So that's a very clear statement. By the way, top down and bottom up. Everyone supported.

Everyone understand that. I I didn't need to explain why we need to fix vulnerabilities. That's not an an argue anymore, but we needed to find the right solutions that are really accelerate the day-to-day and close to the developer.

It's a totally democratization and it's not easy. I was just saying I I totally understand, but it it's not relevant if it's close to the code or at the the latest stage is important because of governance and control. Great. But I'm

saying if the amount of code is growing even at in my IDE and I still have this amount of vulnerabilities is still a challenge today to ask the developers to fix this amount of vulnerabilities. I'm just saying it's a it's unsolved

I agree problem and so I mean thinking about that sort of attestation towards the end.

Yeah. Um was that was the use of sort of AI functionality within a product or use of you know external AI providers within a product was that something that those attestations um you also looked for or was this you

know this became a thing over the last year was that something also search for you know which AI providers are using are they suddenly using a new AI provider we weren't expecting is that is that something that was included as well

yeah for sure okay yeah by the way I do see a a trend that uh just reminds me uh a bad practice from the past where the CISOs prioritized AI security at the runtime before they do that in the development practice

fail the organization that doesn't work this way I I totally agree we're preaching to the choir but I'm just saying that this is why this gentleman developed an AI firewall and there is a demand.

I didn't develop it. All those people are okay here. But what I'm trying to say is that we see that in AI specifically budgets are shift or or prioritize budget towards runtime AI security versus the development. Uh um and yeah, so we need

to do a better job on on educating the board. Yeah. And there are tools today to inventory everything that you you're using in your code for AI and govern it early as you said in the development process.

Yes. When you when you talk about runtime AI security is that sort of trying to you know monitor traffic as it as it's I don't know leaving the application or going to going out to an AI provider. It's all all the um

ORS block 10 for AI in prompt injection and other types of attacks. It can be um elucination or data excfiltration and others. Um but yes, I I developed an agent in my application. I'm running it now um in production and an attacker

tried to exploit it. So I need a protection. By the way, this is again in the risk assessment that we provide to our customers. It's from the code and runtime. Uh we we must prove an end to end view from code to runtime on AP and

AI security. Mhm. Okay. Okay. And then I guess to expand expand the question slightly, um I can't believe mentioned this to me, but you we talked about it in engineering up until now, but you know, how do you how do you

deal with the risk of your marketing team going and spinning up a website with Lord knows what personal information about people in base 44 or something like where how does visibility and how how does control come into that?

We I I I we literally had a call with the CISO today on on this topic. It's it's a it's a huge challenge. Um and they can on board to base uh base 44 by themselves. They don't need an approval

from the seesaw and they can run it and they can connect base to an Excel spreadsheet that has all the PII but you know personal information and identify information of these customers of their customers. It's a big problem. I don't

think that it's like per se application security problem. It's it's yeah it's you need to have a visibility to which uh you know solutions and services and SAS providers you use um and then make sure that it's a data problem that you

need to make sure that you do not provide uh sensitive data to this five coding tools.

I don't love the question because you're using the word prevent prevent the marketing team from doing something. I actually think that is one of the protective protective development. Um because how amazing is it that anyone in a company whether they're marketing,

they're product management, design, coms, you know, HR, legal, finance, and of course engineers can now use AI tools um to build software themselves. like we have so so long talked about democratization uh in roles um you know in countries

around the world depending on on on on your educational background or not right that now you can in almost any human language start building a marketing web page and so I think that we should acknowledge that that's an amazing thing

and I actually believe that the holes between product management engineering design will collapse much more than they have before this will no longer be the side where the PM writes the spec and the designer builds a Figma and the

engineer builds code, they're all going to be building together and they have specializations in each direction of the vend diagram, but the in the middle is much bigger. And so we have to think about protection and you know the

answers are actually not that different from probably your over meet up 5 years ago which is we got to skill people. We got to tell them about the the over top 10 and and the the risk and you know how

to manage data within GDPR and what have you. And we have to give them the right tools and the right platforms. And we have to treat a marketing employee with the same decency as an engineer which is like you are grown up employee of our

company and we expect you to follow the processes and the policies but we also allow you uh to contribute to the value of the company. I think that is crucial in in the mindset that in the shift that

AI is going to bring to our companies. There is no longer all these silos where uh marketing is not allowed to ship software. Yes, they should be as long as they're following the same process and go through the same trainings and so on.

So you see an executive that wants to see business velocity, business growth. Uh but I would just say that on the other hand, I would encourage these uh SAS providers like Bass to embed security inside the product by design for every prompt. Exactly.

And unfortunately they are able to grow faster than then their time that they need to invest to secure the prompts and the code and all this.

Um so it needs to come from the customer saying yes I'm buying but show me that it will be secured and it's hard like it's hard to implement.

Yeah I think I think that's definitely a part of it and I think the other part of it which is sort of why I didn't like why this question is a bit tricky in the first place. I think if you this is sort

of an an evergreen truth but I think it's even more relevant now that you know security we're sort of trying to put more blockers in place more sort of gates in place and more sort of slow down then that's yeah that's the thing

that's going to make people want to use an external platform to begin with. If you if the if the processes I guess are you know to help and to assist and to support then you know people aren't hiding under the table every time

security coming in I think then it's more likely that either they're going to use a platform that's sort of within your area of knowledge or alternative they're going to come you know they're going to come to the security team

involves a security team um that does kind of bring us into the next question which is around by the way sorry before you move on until then the breach in base 44 before and when it will happen they will be they

will invest resources to secure um their platform so so we wait until bridge then we tell people to use it we see other platform like that like salesforce yeah I'm just sharing um huge lot of nights I spent the sales force

issue but uh we see platforms like base but platforms I'm not saying like specific sound solutions but platforms that truly making a huge difference in the companies. They will have to invest in security by design and we saw what

happened uh to Salesforce after Salesfor and and several other vulnerabilities and and huge attacks. So, they're going to do that. They're going to invest in that uh for better or for worse, but they will do that.

Um yeah, I think uh certainly we're hoping to see more of that. I guess we don't see as much of it as we like and not as much of it with organizations that haven't been breached yet as well

because application security teams and and product security teams they cannot think about all the features all the use cases and develop it by their own it's mass so it's really and expertise by those solutions yeah okay so I guess yeah so the next

thing I want to talk about was around you where do you seen apps teams adapt how you've seen them adapt over the last year or so sort of you keep up with this sort of pace of development and again

try and avoid being blockers but or you know work alongside run alongside development team here. What what good examples have you seen of that over the last over the last year or so?

Wow, I got two microphones now. [laughter] I would say that uh what we've seen in in our organization is uh I'm [clears throat] going to go back to your your happy days of oh I'm a new developer and it's three weeks in and

I've got all this tech debt. Well, if you can imagine Aami um and its foundation in regards to our content delivery network and you think about some of the the legacy architectures that we've had and you think about some

of the additional acquisitions that we've made in regards to additional platforms and the complexities that that brought. Um we see this as a a huge gamecher in regards to allowing us to take these disperate tech stacks and really come to a common ground. Going

back to your point of I don't need to have been here for 25 years to understand all the complexities in this text act. I now have a democratized capability that allows me to pull together and leverage all of these what

used to be disperate assets and now bring them together to enhance security. So I think that's really the game changer that we're seeing. The other thing I would say going back to the original question, um, let's not sugarcoat or forget history. There's

going to be breaches all day long. So, I don't know if you remember like 15 years ago when you got your first breach notification cuz your healthcare provider decided to outsource the processing payment. The next thing you knew, every day, every month, you got

another letter from somebody else. Just because we're using AI doesn't mean that we're not going to have those same risk situations from a breach. Furthermore, the reason why organizations choose to actually utilize thirdparty SAS providers is a risk conversation and is

a scenario where they actually defer the risk from themselves in certain areas of their business to a third party. So, we'll continue to see a rise in breaches forever. I don't think that that's going to change. Yeah. Well, I think that the

velocity of the breaches that we will see in the next year or so will will increase because when we measure the amount of APIs in code that are being developed, when we measure the open-source dependencies, when we measure the amount of new technologies

that are being adopted in the code bases across all of our all of our customers, it's growing in in numbers that that You can as a human you cannot handle and map the attack surface and assess the risk

of these um you know components that are introduced and eventually it will come to production without review and then you have like we're so many years in security you have defense in depth so you have a layer on the development side

that protect you have the layer um at the runtime and you do more you know pent testing or other stuff to identify but I do think that we will see an increase in in the attack in the wild

because of this expansion of the development velocity but sometimes we're saying development velocity as a generic title without actually understanding what does it mean it's not just the lines of code okay it's entry point to your application that an

attacker can exploit. So if the number of entries is growing um then you have a bigger attack surface and and you don't even know it as as we said earlier because the developers themselves doesn't understand software architecture. Okay. Um, so yeah. So any

other It needs to wrap it up in a sec, but it's just like any other good examples you've seen or sort of um ways you've seen outside teams sort of work well to to deal with this acceleration.

As as a one that didn't sleep for the last 10 years because of that and other vulnerabilities, I want to bring some optimistic side here. I think it's amazing opportunity for app6 team to be able to put your knowledge your

expertise into agent. Don't be afraid of it. It's what replace you so fast. Put your knowledge your expertise the things that you know the most the best from everyone that is sitting here. I'm sure that you can put his knowledge over

agent and slowly create tools that will help you and bring value to the company.

I think that's will be that what we're going to see uh a revolution not everything will be replaced by us but we will extend the agent they will be our superpower so I'm I'm very optimistic I I'm seeing I'm seeing a lot of early

stage startups that are coming to that point and trying to solve uh features okay it will be a features uh soon and we're going to see a lot of acquisitions very interesting cyber security acquisitions in Israel that are really

want to help us to to make our life easier. I will also be an optimistic for a second and a second for a second. Um and I would say that AI created an opportunity for prevention and use all

the knowledge that we have as synopsic engineers to help or guard the agents the AI coding agents guard them from generating non-compliant and vulnerable code and I think this will be the next wave of ABSSE um good luck to all of us

protecting our applications. Some final thoughts from Thomas. I mean I I think I already said most of the things um you know optimism has fallen a lot. I think you know if you look into bigger organizations the challenges that apps teams often have is

that they're seen as the police and bureaucrats or you know in the US we would say the DMV it's the where you go wait for hours to get you know your driver's license delivered and that's what we don't want right like we don't

want a part of our organization to be seen as the person the people that slow everybody down and so it's crucial you know for appsac teams um as for many other roles to think about how can AI make their

processes faster? Like we all live under the problem of having way too much work, right? Like raise your hand if your backlog is empty, right? Like and or if you have no ideas anymore what to put into the ice box or or the the the the

new idea um backlog or list or whatever. And and and the same is true for apps.

There's always more work than we can achieve. And so the question really is how can I leverage AI to make my life easier or how can I leverage AI to make my customers lives which is often an internal team easier to interact with

me. Maybe you know I build an internal bot you know has all the apps knowledge and markdown files so I don't have to answer questions in SA channels people can self-s serve these questions. maybe can you know start the process of

engaging with appsac pull requests that where AI helps me to write the pull request so I don't have to you know sit in an hourong meeting to fill out some some spreadsheet or what have you and I think this is where AI can help us

beyond just you know fixing vulnerabilities and overstrip lists and all these kind of things a lot of what slows big companies down is all bureaucracy all the meetings you're sitting in you know the seven pages you have to write and small companies don't

have that until they do right and as such I think That's often the overlooked part of how AI can help us. It's these little tools and the little processes that bring us back to to having joy in in working for our companies. And look,

software engineering is not a production job. It's a creative job, right? Like that's what makes it so hard. That's why all our all projects are always late, right? Because with agile and all the scrum and all that, we still haven't

figured out how to actually measure and estimate how long work takes. And AI hasn't actually changed that either. Um it maybe now you're spending more time figuring out the problem than figuring out the code, but you still don't know

when you're done with your day. and and as such you know let's bring the joy back into appsack into software development and see AI not as this bad thing that replaces our jobs but as this thing that actually brings the joy back

into the jobs the joy that we had as you know 15year-old or 20-year-old or 25-year-old whenever we are in coding bring that joy back into our day-to-day that's you know I think my where my head is every single day and like how can I

make my job as as joyful as possible aspect totally agree with And one idea um following to that is think about how you can give selfservice to developers not just bring more work but also think about self-service how you can give them

an agent that will accelerate their their job. So cuz it's it's mutual responsibility. So yeah. All right. Fantastic. Um we're going to leave it there for now. Um thank you very much to our panelists.

Please round our panelists. [applause] in the ground. We've got an update coming up.