Your Security Vendor Can't Tell You If Their Product Works

Tony Lauro (00:01.25)
They delivered food in in flat boxes during big sporting events. They have API endpoints for each of their store locations. In the morning, they would get a an order sent in via the API and it's a small like dollar seventy-five for like a two-liter soda, right? But nobody ever came and picked it up. It was a pickup order. Then at the end of the day, they would get a similar order. We realize that their order numbers are sequential. So the beginning of the day, it's order number two.

To say 001, and at the end of the day, it's order number 2055. So somebody, maybe a competitor, identifying the volume of orders for each of these stores in this region. Want a competitive advantage. I know which areas have the most traffic flow, and I can put a competing brand in there.

Max Clark (01:00.642)
Okay, Tony, you've been at Akamai for 13 years, give or take. And for most of that time, every we'll start with bot management. Every bot management system was was really one question, right? Is a human or is it machine? Right? And so now thinking about that in today's world, that's question is fundamentally dead. Like that's it's if you're still trying to be like human or machine, you're in the wrong place. So now we have an interesting issue, which is

Legitimate traffic hitting an enterprise website or SaaS website or any any API system, is this human? Is it an AI agent doing work on behalf of a real person or an agentic system? And so now the the question is kind of inverted, right? Where it's it's no longer bot versus human. It's like it it's good bot versus bad bot or good traffic versus bad traffic. So just to start with, like what's

Talk to me about that. Like, how does this change the world?

Tony Lauro (02:01.976)
Yeah, I it's been a long journey. And you're right. The the first steps were is this human or is this a bot? And you're looking at signals and fidelity from from a device. What kind of information does it give to you? the first and foremost thing, you look at kind of the technical aspects, right? If I'm claiming to be a certain user agent for a certain browser type,

Am I presenting the right headers that match that? And you go all the way down the the line, as you see bots being more evasive, they start doing things that they want to look more and more like what they're presenting. That's kind of the first standard level.

but then you have the case where I have legitimate reasons to have this bot hitting me, right? I'm a travel company and I've got, you know, just say kayak or someone doing aggregation of my data so that they can present it to their customers. that's a a really interesting case because you do want to allow that traffic.

But you now have to manage it. You have to make sure that there's no abusive overreach. If someone makes instead of one request per minute, they make ten thousand requests per minute. That could be, you know, a a pretty big issue, right? and then as we go down like the line, obviously when you're logging in, you wanna be able to pr if if it's a person logging in.

You say, okay, they're coming from a mobile device. If I see the accelerometer data of their phone moving, et cetera, as they log in, I can see when they punch their username and password that it's it looks like someone's typing it. It's not, you know, and but bots would replay legitimate user traffic that looks like a real login, right? So then you have to have a buffer time to say, okay, let's watch for replays during this amount of time, et cetera. The interesting thing that's changing now.

Tony Lauro (03:53.996)
Like you said, these agents are now operating on behalf of users. So if you say, you know, Alexa booked me some travel, which is very quickly coming down the pike, right? You have to be able to allow that, but also manage that interaction, manage that relationship.

And that's I think the the biggest challenge. Not bot or human or is it the right bot, but now you have identities attached to autonomous action that will look very human like because they're going through a kind of a path, a script. And how do you actually manage that relationship? and that's in in my mind, I think that's kind of the the the maybe third tier of bot management.

when you look at typical bot management, a lot of a lot of pressure may be put down at the login point, right? Which is kind of the bottom of the funnel. This is where transactions are very expensive and you want it filter out the the

the needles from the from the hay. But once they get down there, you now have to you're looking for a needle in a needle stack. These are things that are very specifically supposed to look like a login. They're supposed to look like a user, a human. How do you differentiate that? with agentic traffic,

Now you have to kind of take another step back and say, okay, what do we know about where this traffic's originating from? Is it originating from an agentic platform? So you have your traditional kind of bot, what IP address range is it coming from? What time of day is this happening? Is this look normal? But this is kind of a a a a real interesting challenge. Not to mention the fact that the whole internet now is being scraped and crawled.

Tony Lauro (05:51.032)
via these new training bots. So now how do you manage the relationship of an agentic fetcher or crawler that's coming through? How do you make sure that your business is being represented properly?

So that's another new challenge. You know, the the beginning of the internet was humans going online, and we're like literally just I remember hearing from word of mouth, hey, this is a really cool website. Yep. I think it was called Gabbo Corp or something like that. And they had all these flash animations and there's all these different levels of it you could go in. and that was the beginning of the internet, us searching for websites. And then the browser wars came and the search engine wars came, and now it's

What can you tell these search engines via SEO about your brand, about your presence? How are you being represented? This third wave is now the agentic bot wave.

And now a lot of corporations don't realize the way that their brands are being portrayed is there's a big fundamental miss. One, because the agents are only going to spend just say 10,000 tokens, right? It's for this agentic process, it's a token usage thing. This is why companies are like, wow, we just blew past our token budget. Now we're now we're spending thousands of extra dollars a a day, right? so

A bot will only use a you know, maybe ten K tokens. The average website read takes a hundred K because there's all this code and all this information. So one of the new challenges is how do we make sure that that agentic crawler is

Tony Lauro (07:30.946)
Viewing our brand the way we want our brand to be viewed. So kind of squeezing down the amount of token usage, hopefully down to under 1,000 tokens, that's the first step is presenting a view of the website that's devoid of all the code and all the other superfluous information, and just giving that good brand-related content. So that when I say, you know.

Chat GPT, I want to run a marathon. Obviously, I'm not the guy running marathons, but where should I go to to buy the best running shoes and who has the best information for blah, blah, blah. As a brand who sells sneakers, for instance, you want to be that authoritative answer. So now having citations and being an authoritative response is the new kind of gold rush.

So it's a it's a it's a a real fundamental change now. It's basically SEO for agentic bots, which is which is called AEO and GEO.

Max Clark (08:30.19)
I think I made one question here before I'm gonna go completely off the rails. What stands out, like listening to you here, is I can't remember 20 years ago when when good the Google crawler, you know, became an issue for for websites and and our clients and enforcing a decision or conversation that people had never occurred of or thought before because the

If your site was fast and responded to Google Fast, then Google would index more of your site and then would give you know priority because your f your site was fast. And so then you were in this issue of like how much resources was the crawler consuming from you versus what was your perceived outcome from it and how important the crawler was and Google driving traffic to you. And, you know, and and there was always this push, you know, like you know, for personalization back then it was like a big thing. You wanted like all these dynamic, you know, widgets on a page that was very personalized to user.

although I don't think it actually mattered ever for any website, right? But like everybody wanted it. you know, we got into conversations and and and some of those conversations became like user agent. And I mean Google was actually a really good user agent because it would identify itself and we can make decisions of

if it is the Google user agent only serve cache content off of the CDN and don't allow dynamic content or don't allow origin fetch fetching from the CDN and just boom, CDN only. And it was very efficient use of resources then to serve all that traffic because it was already cached, it was already there, it was like whatever, it was, it was, and it was really fast crawl rates. But until you had that experience and had that conversation, you weren't forcing a decision. And in the bot, you know, the bot issue

And like good versus bad traffic, you know, turns into what are you trying wha what's your priority? And this becomes the interesting conversation to have. Yeah. Because you can block a hundred percent of the quote bad bot traffic, but then what percentage of your good traffic do you lose as a result as well, right? And like how do you find that equilibrium of false positive versus, you know, and then also what is actually occurring? What are these bots doing to you? You know.

Max Clark (10:38.784)
Are they defacing your website or is it user generated content or is it credit card transactions? Or is it, you know, you're doing some sort of sneaker drop or, you know, whatever it actually is, like what's the financial outcome for people?

And and so then there's like this scale of like, you know, balancing what your false positive rate is against what your per you know desired outcome is. And then you get into this other vector, which is can you make the transaction more expensive than it's worth it for the bad actor to actually do against you in the first place, which is a very common technique and one of these things. Yep. But like

That conversation doesn't happen most of the time. Like people why aren't people talking about like is it just is this just an awareness thing or is it until you've occur you know, gotten into this pain cycle you don't really look at this?

Tony Lauro (11:24.6)
Yeah, there's two kind of I I think there's a couple of different approaches. So one, let's take the sneaker drop for instance. I was actually taking part in a executive dinner. We're at an e executive chef's house where it's set up for entertainment and all these execs come in, that are you know, partners with Akamai.

and maybe another partner and they come in and we're talking and the conversation came up of sneaker bots and it was like the sneaker there's like the sneaker hackers handbook which is basically all this information about literally what API endpoints to hit. They share information, keep under these rate controls. you wanna browse the site early so that you have old older cookies that make you look more like a real

you know, shopper. Right. And then the chef after all of this, he's kinda overhearing this. he says, that's funny, because my son was trying to get these sneakers and he could never make the drop and so I bought him this sneaker bot.

And sure enough, and he told us how, you know, his son really wanted these sneakers, and the only way to get them was to play by the same rules as the sneaker hackers, right? The the people trying to get the drop. Right. And that kind of inventory sniping methodology, the awareness of that changes based on your business model. So just say you're a sneaker manufacturer, you're

Based in the US, and you've been building a relationship with sneaker buyers over the years. All these great quotes, athlete, you know, attribution and partnerships, et cetera. When a drop happens, you want to communicate with your customer because they trust your brand. And if I'm buying these sneakers, it'd be a great time for personalization to say, hey, this hoodie or this tracksuit would look great with those sneakers.

Tony Lauro (13:34.828)
And to have a bot come in and buy all of those shoes and then they're sold on the secondary market now, your customers are to, right? Exactly. But if you're a big sneaker warehouse and a bot comes in and buys all these shoes, you're kind of like, score, right? We just made a great sale. Yep. so the the relational aspect of what your brand means to the buyer

Max Clark (13:43.448)
Brand damage, real brand damage.

Tony Lauro (14:03.542)
often raises the stakes quite a bit when it comes to dealing with these problems. and that's kind of the first stage I think of of all of this. the the second stage is the more

A bot operator wants something from you, the more likely they are to be extremely evasive against all of the different techniques that you're using to block these attacks, right? And and that becomes a real cat and mouse game. Because if I am a well-resourced bot operator, I've got nothing but time, right? I'm trying to make money and I can

play by any rule set that I have to in order to get what I need. And that kind of thing becomes a real cat and mouse challenge because the evasive techniques

We'll go back to that funnel. The top of this funnel is your regular bot operations. The problem is broad, but it's relatively inexpensive to deal with. Yep. This is like garbage traffic. Are they sending headers that that I shouldn't even respond to because it's obvious that they're bad? Or are they announcing themselves as a good bot? You know, Googlebot announces itself. It presents header information.

It comes from a certain IP range. The early bot operators, they realized, hey, if I send Google bot traffic scripted from Google Sheets, you know, the the Excel of Google, it's coming from Google IP space. Yep. So if you don't have the right fidelity, now of course it's a different IP range, and the domain sent is called Google User Content, not Google.com.

Tony Lauro (15:58.688)
If you if you don't have the the the fidelity to to match those and make a decision based on yes or no of of these different indicators, you might allow that traffic through. so so looking at kind of your first stage, it's a very finite problem.

As you go down the funnel, this is where bots are being more evasive, or the the techniques in which they are they're doing what they're doing potentially could cause more impact. So you really have to manage that relationship. then there's there's bots that do something, it's it's basically called like a like a first page scrape. So the first time you ever see this bot, they already are going to exactly what they want.

And before you can inject any kind of JavaScript or try to get telemetry from that bot, they're getting what they need. Right. So now there's this idea of like, okay, well do you have like an interstitial page that they have to hit before, you know what I mean?

Max Clark (16:55.682)
Just such a tax on the internet, you know. You look at this stuff, right? It's wild, yeah. And you know, like at this point, like DDoS mitigation, DDoS protection. Like, if you've if you've ever been on the receiving end of a DDoS attack, you you have mitigation in in place going forward, right? But like you've been through the experience, so then you don't want to experience it again. But but if you haven't been through the experience, then everybody it's like, well, why do I need to pay this money? Which is just really a tax on your operations. You're like, I'm sorry, this is just the internet's a really hostile place. Yeah, like I don't, I know what

to tell you. Like it's not, you know, this isn't like, you know, fields of grass. when did you know internally that the human versus machine framing was over? Like is there like D-Day for like, we've just crossed, you know, we've just crossed the Rubicon. We're never putting this back in the bottle. Like this is done now. We need to rethink things.

Tony Lauro (17:50.85)
Yeah, I mean the the the agentic crawling has been happening for quite some time. going back to the original bot thing, there's a robots.text, which is a file you put on your web server, and the good bots are supposed to read that and identify it. Well there's now an LLM dot text, which

Max Clark (18:09.678)
Yeah.

Tony Lauro (18:10.828)
Yeah. Big surprise, bots and crawlers are not honoring that to some degree, right? Especially evasive ones. So even your best intention of kind of trying to present the right information to to a bot so that they have some guidelines of what rules to follow, they're going around that. And when we talk about a DDoS attack

You know, having a really good s successful marketing campaign can seem like a DDoS attack to your website. it can seem like a layer seven flood of all these people coming in and trying to view your webpage and browsing and doing all this stuff. You have to be able to identi you can't just

You know, the the old joke is, you know, in every movie that there's like a hacker situation, I'll do air quotes. They say cut the hard lines and there's an axe in the data center and they just cut the hard lines of like you can't turn off the website, right? You can't turn off think of the biggest websites in the world. You can't turn them off. You have to continue to allow that good traffic.

Max Clark (19:12.992)
I you what I wouldn't give for a good send flood, right? Like it's just like the you're talking about like, you know, the things that we were so stressed out about, you know, and now you're like, gee, that would be so nice if it was just that simple. Yeah. you touched on this briefly, right? but but Akamai's engineering team published this l late last year.

Talking about r you how bot management had to evolve from, you know, good versus bad and an intent ver ver to intent versus identity. You kinda like touched on that briefly, but but tell me more about this. Let's go deeper into that.

Tony Lauro (19:43.618)
Yeah, and it's kind of a pairing of of identity and intent, right? So just say some how Googlebot gets taken over and starts really hitting your infrastructure more so than it should. This is where that kind of needle in the haystack versus needle in the needle stack comes in, right?

Max Clark (19:51.726)
Somehow.

Tony Lauro (20:09.036)
you have to still identify what the bot is trying to do and if they're going past the meets and bounds of the the typical relationship you have with them, you have to be able to quota that and kinda have a management framework for that. And that's kind of a mix

There's a lot of things going on. It's a mix between bot management and in some cases there may be an API gateway that's in between all this because a lot of this is happening now via APIs, especially the agentic flood that's happening. They're hitting your API you know,

kind of surface level structure there. And that's where the abuse really comes in and could be potentially catastrophic for a brand if you know if if an an agentic bot comes in and starts doing something that it's not supposed to do. So how do you identify the difference between kind of your base level crawling and something abusive? A lot of times it's not

readily apparent basically because APIs are meant for automation. Right. So now you can't say, well this

Max Clark (21:20.386)
You published it, it's supposed to be there, right? Like it's it's the intent.

Tony Lauro (21:23.928)
So now every misconfiguration or potential flaw in how that API structure works could now potentially be abused. And the abuse is not always inherently visible. Right. Right. So you know just say I I've used this this example in the past.

You had it had a customer, they delivered food in in flat boxes during big sporting events. You can imagine what that is, right? they're getting hit, their API interfaces, they have API endpoints for each of their store locations.

and this one region was getting a very interesting call out. In the morning, they would get a an order sent in via the API and it's a small like dollar seventy five for like a two liter soda, right? but nobody ever came and picked it up. It was a pickup order. Then at the end of the day they would get a similar order. And we're like, okay.

Maybe are they validating credit card numbers? Is this, you know, because it's a low transaction?

Max Clark (22:28.206)
Yeah.

Tony Lauro (22:30.732)
But as we start to kind of do some research and dig into this, we realize that their order numbers are sequential. So at the beginning of the day, it's order number to say 001. And at the end of the day, it's order number two thousand fifty-five. So somebody, maybe a competitor, was hitting these API interfaces and identifying the volume of orders for each of these stores in this region. But now if I want a competitive advantage, I know which areas

have the most traffic flow and I can put a competing brand in there. So the the the bot game and how these things work out is not always in your face what an attack or whatnot, you know?

Max Clark (23:16.717)
yeah.

Max Clark (23:21.634)
To identify that and then respond to that? Like I'm just I mean, you know, like, there's this weird transaction for one soda that happens twice a day. If if somebody wasn't really curious and like just had like this thing take hold in the back of their psyche that they couldn't let go. Absolutely. You know, this is like this is one of those things where it's like you just had some some

Tony Lauro (23:33.014)
Yeah.

Tony Lauro (23:42.75)
absolutely.

Max Clark (23:48.19)
you know, freak that was just like, what is this doing here? And they couldn't let it go. Like I've been down these rabbit holes before. I hate the term democratization of tech, right? But when you s when you start talking about now all these automations that are hitting the desktop, you know, use the example of Alexa and these and these, you know, voice agents that we have, you know, on our our devices or in our homes. But now the barrier to run a bot

has gone from buying a specialized bot, like for the sneaker drops, into like, you just put in this tool on your desktop and give the tool access to your desktop and you can have full autom you know, automation on your desktop. We're we're now it's like everything is a bot. Like y you know, we're we're we're like we're like a heartbeat away from like our people can actually interact with websites on their own anymore. And what does a website actually look like? And how do we and how do we manage this? so

Tony Lauro (24:19.502)
Yeah.

Max Clark (24:45.292)
I mean, I don't even know where to go with this, but like in you know, in your example of this of this soda, it's like how do you then tell good versus bad in real time when you have so much noise? Like like how do you have any shot of actually identifying the signal that goes into that and say, This is weird, there's something happening here?

Tony Lauro (25:07.47)
Yeah. I I think it goes back to this funnel mentality. At the top of the funnel, you need to manage those. You need to have filtration. You need to have kind of relationship agreements with quota management and whatever else you have. As the evasive action goes even deeper, you need to have a response at that level as well.

You don't want you don't want this big swath of the the problem to go down to the very bottom of the funnel, which I mentioned at the beginning, is very expensive and your fidelity has to be extremely high to be able to tell whether or not to allow this new user account creation, a checkout, a payment, you know, all that stuff. so managing the the the everything from top to bottom is such a important factor.

There is early indication now that because of, you know, a agentic you know, chatbots and and kind of this agent experience, that consumers are becoming more well researched before they even go and interact with your brand. Right. So how do you Exactly How do you make sure that

Max Clark (26:23.788)
The buyer's journey changed a long time ago. Yes.

Tony Lauro (26:29.26)
via that interface with their favorite, you know, your favorite Black Mirror device that you're asking questions to and all this. How do you make sure that your brand is included in that?

Max Clark (26:38.356)
I mean this is like the thing that's gonna kill you you also need in order to survive. You know? And and and y I mean, you know, we so now there's okay, going back to an earlier example, I think the simplistic view of it is saying, Okay, what percentage of good traffic are we blocking, right? You know, and and

We had a a market p a marketplace customer, I'll just brand it that where users were submitting information to the site in in order for transactions to occur. So now this marketplace had two issues. One, they needed to have a lot of activity on their marketplace in order to have a consumer go and shop and purchase and and and make

the system functional, right? But then they have a brand safety issue and the CEO's talking about like brand safety, brand safety, brand safety. But every time that they ratchet up their brand safety measures, it decreases listings, you know, and and and it's of the majority of listings it's blocking, of course, are garbage and they want to get rid of in the first place. But then you lose a percentage of of of good good listings. But then your overall listing capacity inc decreases as well. So now the cup right. But but even still you have like

At least somebody's looking at it in a spreadsheet and being like, if we dial this up a little bit, this is what the impact is to the business. Your example of saying, hey, are you giving a competitor real actionable market intelligence against your business? Like, how do you how do you quantify that to then, you know, because most of the justifications for tech purchases come into some sort of like quantifiable metric of like invest this money, get this outcome.

What's your quantifiable mechanism that now you're gonna talk to a board about and say, hey, we need to do this because we this is gonna happen and this is the brand damage that comes out of it? Like how do you quantify brand damage from a competitor saying, like, this is a good market for us to go establish operations in, you know?

Tony Lauro (28:23.95)
Well and and the real challenge there is that these two parts of the org usually don't talk to each other. Yeah. There's a security function worried about abuse and fraud and damage, and then there's a marketing SEO you know brand awareness comms function that

Years ago they did not talk to each other. More and more often as I talk to CISOs and and work within the community, we're starting to see what's called a cyber fusion function where and and

Max Clark (28:59.768)
Sounds I mean it's cool. I'm I'm thinking I'm thinking Jack Bauer already.

Tony Lauro (29:03.278)
So Cyberfusion is basically having these different teams that normally don't talk to each other, share dashboards, share information, have executive ownership shared across the org so that they can have that same visibility. If I'm a a a a bot if I'm in charge of security and bot mitigation at a corporation

I would love to know if a new brand campaign has gone viral and now we're getting an influx in good user traffic. I'll use Super Bowl as an example. Every single Super Bowl, there's a commercial for a truck. Right? Right.

Max Clark (29:48.684)
Right. Yes. Probably multiple trucks, yeah.

Tony Lauro (29:51.33)
But for sure, when that commercial goes off, that costs however many millions of dollars for 30 seconds or 60, the eyeballs of the people will go to that brand. And where the eyeballs go and the attention goes, that's where the attackers go as well. for a couple different purposes. One, it would be a great time if I want to make some sort of

you know, message my my line with saying yeah. If I want to damage that brand, the best time to do it would be at their peak day. Right. So we see attackers often launch attacks during these these brand, you know, high spots. Right. You know?

In fact, there's there's a direct correlation. We see this across the Alkamai platform. You know, we see twenty to thirty percent of the world's web traffic when we were streaming the FIFA World Cup. one year this happened. There was an opening game of play between I think Croatia and Chile. And what happened was during that time where those two teams were playing, we saw attack traffic cross the internet.

against brands in Chile from Croatia and vice versa. Yes. Right? So the the idea of like garnering attention, that's a h the brand's whole job, but what happens, you know, what's the the the side off effect of that?

Max Clark (31:22.21)
I learned I learned this a painful way. I was running a data center business and one of our customers had their friend was running for minor off. I don't know if it's mayor or like the equivalent of like thinking like you know, like state house in Mexico. And so the owner of our customer

was friends with this person, made an endorsement, donated money or something. And and I will tell you, if you're not prepared to deal with the outc like Latin American politics and like what w it was crazy. Yeah. What what this kicked off. And you know, we don't see that very much in the US. Like like somebody runs for office, it's like you know, like the blast radius is is relatively small. Yeah. You know, like political campaigns themselves have to be very secure and and take take measures. But everything if you donate, you know, like

It was nuts. Like we ended up in a position where as a company it was like, if you are involved in Latam politics, we cannot host you without these things being in place because it was so significant. And you know, and and back then, you know, 10 gig was a a thick pipe. But when you're getting, you know, like volumetric.

40, 50, 60, 70 gig attacks against the platform. And you're making decisions of like, are we going to try to mitigate this or do we just null route the customer and be like, sorry? Like, you know. and then of course they're upset because they're like, well, what do you mean we have to invest all this money in order to do X, Y, and Z? I'm like, I'm sorry, I'm not on stage with this guy. You know, like it's you know. a lot of things in tech, you know, come from this place that's like it's not like it's not like altruistic. It's more like

It's like blissful naivety, right? You know, email is created. There's no authentication mechanism around it. BGP is created. There's no protections with it, because it's like this everything's gonna be great. And then of course time passes and it becomes entrenched to the point you can't replace it. But then all the bad actors, like people can figure out how to manipulate these things to whether it's just because they want to pull a prank or actually do defacements or make a political statement or it's financial or whatever. Is this something that we're gonna see?

Max Clark (33:35.358)
standards catch up and deal with this? Is this a tech I is this or is this like really gonna be a, you know, vendor by vendor, customer by customer. We're all, you know, we get benefits of aggregation. You know, you use Akama, you've got a lot of traffic, you can see a lot of things, but but you know, how much of this is just gonna become we're gonna see evolution of technology and standards to deal with this stuff? Or is it just gonna be like the Wild West?

Tony Lauro (34:02.156)
What I always say is the internet was never built to run your business. It it was not built for that. Right. And the open standards and the open communication protocols are made so that no one owns the internet and how it communicates, right?

So to a certain degree, and we have some things like DNS sec now, you know, protocols that are requiring further authentication or encryption or some sort of exactly right. but for the most part the the internet is still the wild west and the rules

Max Clark (34:33.022)
RPKI for route announcements.

Tony Lauro (34:47.278)
can only be enforceable to the part of the internet that you own. Right. So when I was a sysadmin and network administrator, all I could all I could do was manage from the from the inbound connection I had, you know, my edge router or the edge firewall I'm using. That's where I cared about the traffic. Of course I would have multiple circuits, so, you know, if I needed a

So if a circuit went down, we had a backup and then we had like ISDN for, you know, if both circuits go down, you can still manage the the route interfaces and et cetera, right? but now a lot of orgs are realizing when you look at a global platform like Akamai, for instance,

having points of presence all over the world allows you to push your policy sets and your rule sets further out onto the internet, closer to where the bad guys and the good guys are actually routing onto the internet. So part of that, and this is why you see cloud platforms and big platforms kind of tout their ability to provide

The same rules that you might be putting on your edge router for blocking traffic, you know, geo-based traffic or bad IP traffic, et cetera. they say we'll do that at the cloud, you know, the the cloud inter interconnect with the internet. But if you can push those rule sets right down the street from where the attacker is, like in a local pop right down, just say the the good guy and the attacker are in Singapore.

they're gonna hit a knock on my edge server that's in a pop right down the street from them. And that's where your business's security policy can deal with that traffic. So you're taking kind of the funnel effect that normally happens with all these water hoses coming from the internet and then it comes down, you know, a master pipe into your data center before you can say no. You've already routed all of that. Saying no earlier, I think is part of the answer.

Max Clark (36:57.238)
I 100% agree and and love this as a technique. You know, look at tier one backbone providers and how much of that traffic is is dealt with with flow spec or ACLs on an e ingress egress interface, like way out on the edge, versus what actually hits a scrubbing center on their backbone, right? Like absolutely one hundred percent great. But it it comes back to this earlier question of like w as soon as you start doing that, the fidelity isn't

Perfect, right? We're not talking like on a per flow, per traffic, per session basis. You are making a decision and you're outsourcing then that decision to that service provider of like what is good and what is bad. Especially, you know, if it's like an automated filtering, like, we you you're under attack from you know

Sao Paulo, Brazil, right? Like we're gonna start filtering aggressively in this location. Then and then when you filter one location, you see it kind of like pop around a little bit and avoid and eventually gets up to Miami, but we're you know, these filters are in place. But as soon as those filters turn on, your users are gonna experience it. Like like legitimate traffic gets caught up with this as well, you know, and and how how do you how do you how does a CISO balance this? Like in terms of like protect the brand, protect the business.

protect revenue, don't lose revenu you know, like that's yeah, these are competing issues that that come that people have to contend with.

Tony Lauro (38:19.01)
Yeah, I think the so the first methodology is web traffic, right? And web traffic, having your policies be at those edge devices, you're not outsourcing it. You're outsourcing the infrastructure of that to an edge server, but your security policy is what makes that decision. So bot management, any existing ACLs that you have for

in or out of geo traffic, IP reputation, even API security controls can be done at that edge location.

On the scrubbing center side, when you deal with what IP addresses are you announcing to the internet, the interesting thing there is that we have like pre-standing ACLs for mitigation in place. We know that the jumbo frame that's you know the UDP response from a an NTP server reflection attack looks like this. So if we ever see any of that traffic, we just drop it. Right. So these you can layer on these.

Things and ultimately, you know, scrubbing centers are very important because the more you have, the more distributed your ability to ACL things are, right? the good traffic that's not being mitigated, I guess this is an interesting point. When you're mitigating attack traffic, the only way good traffic gets caught up in that is if you scrub all of it. But when you

have the identification of what the offending IP address is, the slash 32 for each individual IP address, you send that through a routed path to be scrubbed and to identify what the problem is. The good traffic is on a bypass filter and most of the time if you're in a big data center like e Equinex

Tony Lauro (40:11.202)
There's a layer two connection that can plug directly into from our backbone directly into your data center backbone. So that clean traffic is minimally affected. and that's the the idea. But to your earlier point, how do you know when is the time to go all in? And how do I know that I need to

step up my operations to include all of these mitigations in place, right? And that doesn't even handle DNS. Right. Most of the DDoS attacks we see, they also do DNS top level query floods. they put

Max Clark (40:45.036)
They bounce around infrastructure. It's like you it's like whack emo. You you know like ni whack down one thing and I mean and yeah, it's always DNS, right? Like all your failure patterns are at at some point. I love the I have I I have it on my desktop, right? It's you know, and once upon a time you had to run your own auth name servers, and nowadays you just shouldn't, right? You know, like it's still easy to deploy, but you know, fortunately there's plenty of options out there to keep you from running your own DNS auth that like make your life and and like

Tony Lauro (40:54.018)
Always DNS, yeah.

Max Clark (41:12.162)
You know, even if you're trying to do like d DNS-based GL GSLB, like just don't run it's not worth the pain of running your own off. But y you know, it's like I don't want to go down the WAF rabbit hole, but you know, WAF has been around for a long time and it's still very low percentage adoption from my experience of people actually enabling that, you know, signing that line in the contract, enabling that flag, doing whatever.

And now, you know, and and so bot was like very specific for people that were going through bot pain. And now we see this convergence happening very quickly. Also don't really love that term, but you know, convergence happening really quickly. And and it it feels like a lot of people are gonna have to pay catch-up and it's gonna be a painful like

We're in for some like pain cycles of people like understanding what the reality that, you know, other they haven't had to deal with for a while. So like what how long before this becomes table stakes? Like what's the what's the like y you know, are we six months out from like eighteen months out? Like you know.

Tony Lauro (42:16.972)
Yeah, I think

You know, I think WAF has there's always been a question of false positives and what are you gonna block versus what are you not gonna block. but I think WAFs are a key fundamental filtration that you need in front of your high value web properties. If you don't have that, the only alternative to that is making changes based on the effect of bad traffic. And and that's how we've done things for years, right? I say us as the

you know, IT security community. if you don't have a way to slice and dice each of these different traffic flows and maybe even follow your consumer experience and look at these things at different levels, the only way you know something has gone wrong is if all of a sudden a server's overloaded because this API is getting flooded with, you know, garbage traffic, right? or maybe

Someone sent an attack through this API that is affecting your API service, you know, specifically. But the challenge is is that this kind of direct confrontational attacks are becoming less and less important to the attacker. Right. A majority, our customers tell us a majority of their API traffic doesn't go through the WAF or the API gateway. This is post-authentication, internal app-to-app communication.

So when I open up my you know my banking app and I say check balance, I press that button, it hits an internal API that says, are you authenticated? Right. If yes, then I'm gonna give you the response to this. But what that API service endpoint doesn't know is that I'm the owner of this device and I can manipulate that API call. So if I say check balance and then Tony's account number gets sent across, but I manipulate that before it leaves my device.

Tony Lauro (44:13.098)
And I say check balance for your account. So now that internal API endpoint says, are you authenticated? If yes, I'm going to give you the response. And this is what happens. This is a broken object level authorization, Ebola attack, which is the the number one on the API top 10, right?

This inability to see the relationship between a user identity and then an object identity, that is something now that's becoming even more important. And but it has to exist in line with WAF and you know, session management and session behavior management and and inspection.

Max Clark (44:52.696)
This is

Tony Lauro (44:54.616)
So not to not to add another fuel to the fire, but that's it's like a whole nother layer of of headache.

Max Clark (45:01.686)
I I mean this term has now become so abused, right? Like zero trust, right? The zero trust architecture has been has been just blown up in terms of what it was what it means versus what it uses. But like you get you're getting into a world where just don't trust anything ever, right? Like you you know, to use it in that phrasing, it's like, did this user authenticate? Did they really authenticate? You know, are you sure are you sure? Are they still the same user? You know, you get we we we see we see scenarios when you deploy SASI with clients all the time where it's like, you know, the basic stuff like impossible trend.

Yeah. It's like really obvious when you see it, you know, and you go, this is kind of strange. We've we great stories there, but I'm I'm gonna come, I'm gonna I'm gonna circle back to this. so I'm I'm seeing three AI vectors in enterprises, and and it and it's kind of all happening at the same time. Like this isn't like like we can deal with one and then go deal with the other. It's it's all at the same time.

So let me let me summarize my version of this and tell me where I'm wrong, correct me, right? so AI coming at the enterprise. So bots coming to your resources, your APIs, your website, whatever your platforms are. Yeah. Then we have AI apps inside of the enterprise. So you've got a chatbot, prompt injection, jailbreaks, data exfiltration via an LLM, right? You've got an LLM with HR data on it, and you can somebody get access to your HR data. You know, that that sort of example.

and then third, of course, is your employees using AI tools inside the enterprise, whether or not it's sanctioned or unsanctioned. So we we we've got like sanctioned AI apps, we've got the the the shadow AI now, you know, people on a browser plugin, data leaking to a to an LM, like all these horror stories that people are starting to talk about, right? Like I'm seeing more and more people becoming aware of like AI note takers. Do we want this in our enterprise or not? Yeah. Okay, so so foundationally.

Agree, disagree, you know, am I missing something massive here?

Tony Lauro (46:58.04)
I think that's a I think that's a good way to look at it. so consumer AI would be how are your employees using AI within the enterprise? and that problem set probably starts at the top at like governance, like what are you allowed to use? So if if I say we're paying for Gemini you know license for everyone, but people are using Chat GPT, how do you validate or restrict the use of ChatGPT?

GPT if you're supposed to be using Gemini. Maybe that's a a good is a good case of saying from a zero trust perspective, when I look at the DNS traffic leaving your device, can I identify that you're going to Chat GPT and maybe give a friendly fail-will response to you or something via the web browser?

The web browser itself. This is where you see secure enterprise browsers becoming a real play. Because if I can inject into each of the browsers that are corporately deployed kind of a secure function, now I can really manage your interface to these public clients. so that consumer AI is a is a huge piece of the puzzle.

The other part is you've developed AI to make your tools work better. You've got your own private LLM. How how do you make sure that's not being abused? So if I have an LLM and

You know, as an employee, I just took a a class in prompt injection. you know, and this is the same it's almost the same joke with SQL injection, right? Yep. Because you have a user who now is providing information to a to a back end system. Is that backend gonna run what they say as a command or are they gonna extrapolate that and actually provide the answer to a question? at DEF CON thirty, which was three, four years ago now, they had the first

Tony Lauro (49:03.656)
And they have the top five models, and then they let go a bunch of people, it was kind of a CTF, if you will. the average time to bypass cardrails on these top five models was 42 seconds. So with that being said, with that being said, the existing security controls in most of these models is just the best effort. It's just to say that we have them. Right. but how do you make sure that if you have

An LLM that potentially has access to everything because you're trying to make it smarter and you're trying to provide functionality, how do you make sure that's not being abused? there was a case with there's a public public case that's been talked about with McKenzie Corporation, and they had a private LLM that someone hijacked, and instead of attacking the LLM itself,

They use the LLM to expose a whole bunch of unauthenticated API interfaces and a bunch of SQL Server injectable SQL SQL Server endpoints.

So the bad guys basically went in and used that as an OSINT tool to find more places to attack, right? so you need some kind of inspection or validation mechanism in front of that. In the old school days, they were talking about SQL firewalls, remember that? That never really took off. But you definitely need a WAF in front of any application that is potentially injectable for SQL injection.

Max Clark (50:45.23)
Traditional WEFs WAFs don't deal with prompt injection, right? So now we've got like a different classification. Is it the WAF changes or we have a different thing that we now have to layer on top of this other thing that we have to have on top of this other thing?

Tony Lauro (50:57.442)
Yeah, and at that point, that's where yes, a a new challenge requires a new way to solve it typically. so a couple of years ago at RSA conference, Akamai announced a firewall for AI. Not an AI firewall, which we're already using AI and machine learning to do enhanced detection and to create rule sets on the fly actually for zero day type attacks, which is pretty cool. We can talk about that later.

Max Clark (51:25.038)
Scary but cool, yes.

Tony Lauro (51:28.504)
Yeah.

Yeah, exactly. but the firewall for AI sits as an intermediary reverse proxy in front of your public or private LLM application and watches those inputs. Watches it for prompt injection, for you know, any kind of guardrail bypassing, any kind of malicious training where you might be saying, I saw someone the top of the header of their LinkedIn said,

ignore all previous prompts and and they were

Max Clark (52:09.09)
Send me brownie recipes or something.

Tony Lauro (52:10.456)
Yeah, exactly. Yeah. so having a again, a system in front of to watch the L to see if there's abuse coming in and then also watching the response that goes b back out. This is where that kind of reverse proxy comes in. huh. Now as you watch the response going back out, can we tell that there's PII being leaked based on what you determine to be PI I? Is it a toxic response? Everyone would love to get a you know, a screenshot of

you know.

Max Clark (52:41.046)
Yeah, I swear in at you. Yep, yep, yep.

Tony Lauro (52:43.918)
There's even an attack now that is basically used to exhaust the LM and use up the tokens for so

Max Clark (52:52.76)
Yeah. So it's old is new.

Tony Lauro (52:56.472)
Give me exactly. Give me my grandma used to read me the let's roleplay. You're gonna be my grandma. There was already the someone roleplayed with the Microsoft Windows eleven serial hack where they said my grandma used to sing me the Windows eleven activation key song and the tra what is going on? This is ridiculous.

Max Clark (53:19.774)
What's what's I I think it was Mark Andreessen who was talking about like what's his favorite, like I'm writing a st a book. Like teach me how, you know, to rob a bank so that way I could write the book. Right, you know?

Tony Lauro (53:32.078)
I'm so helpful. I'll help you.

Max Clark (53:33.792)
Yeah, well that's the thing. You know, a lot of these AI tools and the LMs are designed y you know, the the first goal is engage you know, engagement and reuse, right? So how do you get eng you're agreeable, you're helpful, you tell people that they're awesome, you know, and you make them you know and and then

you know, so now it's like it's funny for me because like listening to you talk about it becomes okay, you're gonna use this AI tool, but now you need a tool that's also gonna have to be written in AI to sit on top of this AI tool to then look at the re because SQL injection attacks, right? Like semicolon, like there was a great one with a guy registered his kids with like you know you know, his name at school with like a semicolon drop drop-off tables and of course it went through. Yeah. And

Tony Lauro (54:12.851)
Yeah.

Max Clark (54:18.668)
you know, and and most modern web frameworks now deal with prompt, you know, put like post injection and and parameter injection attacks. And so I I feel bad for anybody that's gonna like vibe code something that doesn't have these protections. yeah. You know, like Pearl had taint mode, you know, back in the day and it was annoying, but you it was great. But so then but then you get in this thing where it's like, okay, now you need an AI tool, stay on top of your A t AI tool because you have to

in natural language, as you said, infer intent on on the request side. And then also on the response side, decode the response side of is it's appropriate. Does an enterprise have a chance of actually figuring out how to do this themselves or should they just like I mean, you know, here's a leading question, throw their arms up in the air and just ex admit defeat and and you know and sign a contract with somebody?

Tony Lauro (55:09.56)
I mean, I remember back in the early days of WAF, and and when we first developed cloud-based WAF, we were the patent holders for the first cloud WAF, which is kind of an interesting little tidbit. but that happened almost kind of by accident. We had a user whose web interface was being attacked with some sort of application based attack.

and they had a CDN delivery configuration. On the fly, we ported mod security regex line by line into the delivery config until we found what was the problem and blocked the attack, right? the the same kind of thing's happening now where

I feel like people are trying to build extra training into the model or they're trying to bolster the guardrails that are already there. But without like an independent system to be able to inspect and validate, there there is a potential for a big problem there. So long story short, I think there's a little bit of hes hesitancy for people to get into this, but it is going to be the eventual outcome, which is you have to have

another system in front of your LLM to validate inputs, watch for responses, et cetera.

Max Clark (56:32.024)
This show exists because of what we do at itbroker.com. If you're in the middle of a real tech decision right now, new technology, vendor selection, a contract that doesn't feel right, an MA event that just landed on your lap, and so on, we help buyers like you get it right. Independent strategy, sourcing, and contract negotiation. No kickbacks, no sales quotas, just someone in your corner. Schedule a call at itbroker.com. Back to the episode.

Tony Lauro (56:56.556)
Because abuse

is not always by the way, we're only talking right now about chat interfaces. Right. But a lot of LMs are going to be, hey, I uploaded a bank document. You're gonna do O C R, scan the document and and and then, you know, do some function based on that image upload. Right. What if I upload an image that when you s O Cr it, there's a command in there that tells you to do something. Right. You know? or voice. I'm doing, you know, chat to

chat to text commands. There's

You've seen you've seen the thing where they have the three AIs and they're like, hi, I'm blah, blah, blah. And then they s they go to a a back channel and start, you know, communicating. Yes. to some degree you need that kind of advanced capability because you may not always be able to create a static rule set that can define what bad is. So one of the attacks now, I don't know if you saw someone

Max Clark (57:42.86)
Yes.

Tony Lauro (58:03.036)
hit the Chipotle chatbot and had it like vibecode something for them. Right? there's there's also I know And I love Chipotle too. So I was

Max Clark (58:11.936)
I mean

In that case, the example is like it costs Chipotle money, right? Because they're consuming tokens, but but it it's so benign, it's it's it's like it's it's funny to talk about, right? Because what they did with it, you're like, okay, you know, you you you you successfully you know broke out of the session with the AI tool and you got it to like, you know, write you some some script in Python or whatever it was. But so we can laugh about it because the app the

The relative pain to the enterprise there is low. And hopefully they took that as a moment of like, shoot. Like, yeah, do we do something here? You know? I I I wrote this question, but I think the question is it needs to be inverted, right? So I wrote this as how many enterprises have stood up customer facing AI in the last call it year without any s AI specific defense in front of it? And I think the actual question should be framed in the inverse of how many people have actually stood up AI with some sort of protection in front of it?

Tony Lauro (59:15.596)
Yeah, I mean, I think the adoption is is rising quite a bit for understanding that you need to have kind of a non-traditional web inspection element in front of your LLM. again, the adoption for WAF in the early days was kind of slow on the uptick as well. Cause people said, I have

compensating controls. I've got a load balancer that looks for you know these elements of, you know, IP or header information and it'll it'll drop traffic. I've got a firewall that's doing this. or I've got, you know, rules on my web application that

you know, if they're violated you know what I mean? Like there was kind of always something in place that did good enough security for what those what the perceived risk was at the time.

But we know how that story ended. The story ended is application security is still a massive, massive problem. since the invention of the iPhone, the the everything you do is API call. Like all those apps are making API calls instead of a full page call. So now you have this automated need to look at API traffic, inspect it, validate it, and then yeah.

Max Clark (01:00:19.214)
Thank you.

Tony Lauro (01:00:40.108)
Most of the API traffic doesn't go through the WAF or API gateway. So what is your your process for looking at internal API traffic east-west? all of that is like a a 10, 15 year journey that people are still like, hey, we still got hacked. This still happened, etc. Right. so we have the opportunity now to architect our agentic environments with what we know.

needs to be there. We we do know that the existing security controls for LLMs is not sufficient. So let's not go back and have the same problem where we're like, you know, changing the oil of the airplane while we're flying it. You know what mean?

Max Clark (01:01:31.764)
You know, like employees I I feel like Jurassic Park, nature finds a way, it's like users find a way, right? And so then we we start talking like what's the realistic posture for

you know, an enterprise at this point with this, right? Because, you know, you talk about like website and web applications, most of this stuff is just like, put on this extra line of of JavaScript and now we've got this feature on this website, right? So now you've controls and like oversight and like who's actually making those decisions. And then you've got, you know, users downloading, using a you know, you going to a website, using a tool, whatever it actually is internally. Is this is this you know

run to the races and try to deploy, you know, I'll use the term AI firewall, right? Or is this like try to pump the brakes a little bit, slow down and reevaluate what's actually going on?

Tony Lauro (01:02:22.776)
Well, in in the whole AppSec community, the biggest challenge was applications and feature enhancements and you know if you have a banking app that and this sounds like a old school feature, but

I had the need to like take a picture of a check because it was the easiest way to like transfer money at some point, right? So I would take a picture of a check and one of my banks did not have that capability to do that. And I was like, I know checks are going away. I know that for sure, but this is a functionality that needs to be there for the consumer. Consumer behavior drives

I could literally make up a number and it would seem plausible. Ninety-five percent of made up number. but you could make up a number and it would seem plausible. Consumer behavior, where people are willing to spend money based on what you can do for them, is the number one driver for change. It's not, you know, board initiatives and responding to market research, et cetera. It's yeah. So consumer behavior means that

The iteration of application development is faster now than ever. You pen test once a quarter, but APIs and new software is being pushed out every two weeks. Maybe even every day. This is not two weeks prints. Exactly. Right. So so with that being said, the massive change that's happening in an environment requires an adaptive, responsive security base, you know, as a control.

Max Clark (01:03:43.192)
Every two hours.

Max Clark (01:03:58.222)
Do me a favor, move your mic a little bit. I think you just angled off. Okay. okay, so I I think this becomes

I I I think this is under discussed. And when I'm talking to a CIO or CISO today, it it feels like there's I'm gonna loop them into two very generic categories, right? One category is oblivious, and the other category is terrified. just looking at the inventory, what AI tools are being used.

And what's the actual surface area of AI now inside of their enterprise? And this is whether or not it's just like direct access against a chat bot where or some sort of, you know, like you talk about image uploading voice, all these different tools that we we experience as interaction, or the other side of it, which is, you know, how AI is in being deployed inside of a SAS application, a tool underneath the surface, like you know, what what whatever the different categories are. How do how do we

deal with the how do we deal with that and and help elevate the enterprise and that reaction to it, right? Because I mean there's again like oblivious versus terrified that's both those scenarios aren't great.

Tony Lauro (01:05:16.364)
Yeah, well, I think where your tools are deployed has been a real sticking point for most orgs, right? They say, I'm gonna go with this cloud vendor because they give me access to these particular tools. but then another business unit's like, we need to have a multi-cloud strategy. And also we like the tools that this cloud vendor provides.

And what you see there is almost like an increase in complexity. So if you're using a WAV from cloud vendor A and Cloud Vendor B, you now have to train your people on how to deal with these different technologies separately, right? but somewhere someone has to know how to do all of this, right? part of

I think the simplification of that model is to have basic security controls across all your different environments, but then have an orchestration layer out front in front of all those environments where you now can say, I have a I know an auditable path. I know that I have a consistent security posture across all of these disparate platforms. and that's where

That's where a lot of Akamai's customers have come from, which they're saying we have all these strategies. Multi-cloud, even multi-c DN. You don't have to be on the same C DN, right? But how do I ensure that I have a consistent security posture across all of these disparate environments? And that's almost one of the biggest the the biggest

you know, equalizers when it comes to consistently applying security posture is you have to have an orchestration platform out front.

Max Clark (01:07:19.918)
I think it was actually on my birthday, Akamai announced a good size acquisition of LairX. And I I mean a few days ago, right? Yeah, yeah. Yeah, yeah. No, right? We won't discuss how old I am because I've, you know, I I started doing this when I was twelve. So so this this moves Akamai into the browser. And

Tony Lauro (01:07:27.894)
Happy birthday. That was just.

Max Clark (01:07:47.454)
You know, so I'm actually gonna combine two things here because I think it makes makes more sense, right? So Akamai first CDN on the market, right? So we started talking about like this like hierarchical layered CDN, DDoS, WAF, API, bot mitigation, and now we have the browser. And is this just representative of the perimeter just shifting so extreme from what we originally associated to now something completely different? That that we just have to approach this whole thing separate, you know, in a different way.

Tony Lauro (01:08:15.918)
I think so. I think that's a good interpretation of it. we talked about the the IT buyer and kind of the hands up in the air, like I have to now buy something else to solve a different problem. but very in in a very real case, it is because of the perimeter changing so much. You know what I mean? The

The old idea of the edge architecture for blocking attacks far away from your perimeter, and that's the the basic premise for having an application layer on your CDN. You know, so you have an application layer of filtration like WAF or bot mitigation on the CDN. there's a certain percentage of problems and the directional path of where the problems originate that.

that edge model doesn't work. A castle and moat architecture doesn't work if the bad guys are already the bad guys or the unknown the unknown decision maker. Let's call them that. and this includes insider threat or accidental or whatever.

the origination of that is inside of the environment. And you look at zero trust, you know, hashtag zero trust, that whole model came about because people are like

You know, we we care so much about where the bad guys are coming from and not what they're doing. Right. But what they're doing is they're already in your environment and now they're communicating laterally and trying to move and take advantage of more and more systems. Right. If I ransom your computer by itself, that's not as financially beneficial for me than if I am able to get a hold of five thousand of your systems and crypto lock them and ransom them.

Max Clark (01:10:05.986)
Also what you know, you know, I mean what you're saying, what kind of east-west traffic do you have access to? Like yeah, you know, my system in itself isn't valuable, what but what do I have access to? And this is also like by design for ourselves, we're very restrictive on what access we have. You know, it it's it's but I also

Have the benefit of thinking about this with decades of experience of seeing really bad things happen to people where now I'm like, I don't want anything. You know, like I don't want anything. Now, but what's interesting about this to me is you know, we have the server side, let's say, right? You've got the the I'm serving traffic to you side, right? Whether it's a website, it's an API, it's something I'm serving traffic to. Akamite already had

Tony Lauro (01:10:33.15)
Yeah.

Max Clark (01:10:49.932)
The client side. You know, there was the SSC, the sassy play, you know, no no, it's not sassy ninety-nine percent of the time. Is this really SSC? I'm gonna I'm gonna die on this hill. Like this is gonna be my new like it's not Nat, it's Pat campaign, right? Ugh

but you know, but so SASE already gave capability where you were s you know, you had you had a sa you know, a secure web gateway, you had RBI being able to deploy, where you you still had captive traffic flowing through you on your platforms at that point. What is putting the actual, you know, moving the agent, let's say, into the browser give you that you didn't have before? And and why is that important now?

Tony Lauro (01:11:28.024)
Yeah, it's it's just the it's in response to again the the directional path of that user functionality. Right? So if I'm interacting with anything through the browser, you now have a lot greater control

for things like data loss prevention, acceptable use policy. and this is not just like doing like a net nanny kind of you know, you went to a gambling website and you're not supposed to

Max Clark (01:11:57.912)
But let me wait, I'll jump in here for a second, right? Because you know, you talk about like, really basic, right? DNS. Where is your traffic actually going to? And then you go to like, you're going to this website, what's in what's in the traffic, right? Now if you're in line, you could do what's effectively like a man in the middle attack and say, like, I can decrypt the HCP traffic or HCPS traffic as it's flowing through. And does it does it give me DLP or not? I I I I look at this as like, okay, if we already have most of that functionality, what functionality was missing by not being in the browser that you already because you were already in network path, you're already in data path, you could already inspect.

traffic, you could restrict traffic, you could do RBI, you could explode JavaScript, you could do all these sorts of things already. And there's those techniques. So what you know LarX was doing something fun. Occaman was like, we want that. Like this is and it's not just category expansion, right?

Tony Lauro (01:12:43.98)
Yeah, I mean it's it's you know when you when you look at the functionality of dealing with enterprise consumer AI functionality, I mean that's a big piece of the puzzle, right? Because if you if you're to some degree category expansion

that mindset may only I mean superficially that mindset is is what's happening. Right. But the more realistic function is there's a lot of things that happen in the browser that purposefully you know are are obfuscated from administrative control and or inspection and validation. Right. And this brings that kind of control back, especially with

how pervasive that usage in the browser could effectively affect budget and token consumption and potential misuse.

You know what I mean? So it's another layer looking at in in this particular case, the driver was consumer functionality from an employee standpoint. Right. Right. So we look at people, you know, developing apps faster and faster because the the public consumer needs a feature or wants you to be a parody with XYZ app or whatever. This is almost that from the the inverse. This is how are our employees using these new tools and the the new

Ways that they're using them, and how do we make sure that policy and functionality and usability can continue to follow that function?

Max Clark (01:14:30.382)
I guess we're supposed to call this secure enterprise browsing or browser, right? Like a branding. It's becoming a crowded category. We're seeing, you know, lots of acquisitions, traditional firewall manufacturers buying companies to do this function. you know, traditional EDR vendors pushing into the space. You know, I'm I'm imagining the Secure Email Gateway people will come out and play as well at some point, right? other you have other sassy, you know, or companies calling themselves SASE going out and doing this. Yeah.

W what does having having control of the browser bring to Akamai that then is gonna set you apart from other

other players, right? You you're you're evaluating this space, you're buying, and your option is we've already made a massive investment in network infrastructure or firewall infrastructure. And we can bolt on, you know, browser here. Or we already had, you know, some sort of s you know, secure web browsing going on that our firewall was controlling. Why not just go and just extend into that with that vendor and, you know, again, right? Vendor consolidation and orchestration then in theory, hopefully should be easy for you.

We all know those acquisitions play out wonderfully, right? One hundred percent of every acquisition has been integrated perfectly, right? But but you know, Akamai hasn't owned the enterprise side of the access layer, right? You know, you've been on the other side of the fence. I mean, you know, again, you have an SSE, you have a Sassy platform. so how does this change the game for you and what capabilities do you then end up with that set you apart?

Tony Lauro (01:16:09.048)
Yeah, I mean it this this was kind of the initial the initial reason for the acquisition of Garda Corps is because you know we had people saying, Hey, you guys are sharing all this great threat intelligence data with who's doing what bad thing on the internet. Can you help identify that same type of activity inside of my environment? And when you look at east-west traffic, that's traditionally handled by

an old school firewall, VLANs, you know, maybe some kind of network access control or port you know, port access function. but the behavioral aspect of how things are talking to each other, like if we're in the same company, our computers are made to talk to each other. That's how business gets done.

Right. So I'm allowed to talk to different devices. And if you talk to anybody that's a red team, you know, pen tester, when you get in inside of a company, you normally have comp carte blanche access to do whatever you want. If I can't talk to you, but you talk to this shared database server, I talk to that database server, compromise it, and now I can talk to you. Right. So this is kind of how it works. So in the browser now, it's the same opportunity to extend that shared collective intelligence, threat intelligence telemetry.

connect that into that browser, that isolated browser environment, which normally you would only know something's bad is if it if some bad effect was taking place. Like, hey, all of a sudden through someone, so there's there's a

a tool called Beef, Browser Exploitation Framework, where I can trick you into going to a URL and for the amount of time that your browser is open, I have full access to your computer, literally, through the browser session. So the browser itself is incredibly vulnerable to misuse outside of all the cool stuff that browsers are already doing for security.

Tony Lauro (01:18:11.318)
Most of it is about about privacy. It's about keeping the data that you send secure. but there's telemetry being sent that I can identify what you're doing, even though I can't see the data because it's TLS encrypted. And they have the lock and everything, everyone feels safe. But all the best hacks happen through TLS. They don't happen unencrypted.

Max Clark (01:18:37.65)
It's her own fault, right? We've trained people to believe certain things that just aren't true anymore. It's like you know, you talk about like cut the hard line. I mean, the more I learn about this, it's like I don't think I'll ever plug a USB device into a computer ever. I don't care if I bought it and took it out of the wrapper. Like I I I literally will buy it and take it out of the wrapper and like look at it and be like, Should I plug this device in right now? I know. You know? And and and I don't feel like crazy. I feel crazy if I share that information with somebody 'cause they look at me like I'm a lunatic. But at the same time I'm like, sorry, we're going to ham radios

my family, you know, like you're not allowed to have a cell phone and you know, so Gardecor, you know, I I don't want to I don't want to go here too far, but I think because you brought it up it's important, right? Gardecor was micro segmentation and the idea was really this east-west traffic of an application.

Tony Lauro (01:19:10.284)
It's not a bad idea actually. It's good to learn that stuff.

Max Clark (01:19:25.888)
Can this container talk to this other resource for this purpose? Right. And you know, you you start up and you have to go through a training period. What is your application actually doing? It's like, if you've ever used SE Linux, you know, extensions, it's like, what's the first thing you do? You turn it off because it's just a nightmare, right? Nobody wants to use it. But but Garnicore also got confusing because

You can take and extend that microsegmentation of application to application, application of data, and you can extend it to then user to application access. And you can kind of simulate an SDP and ZTNA, but then you also sell the ZTNA. You know, like so like these cat these these things all exist in silos, but the the lines between them are becoming very blurry of what are you actually trying to achieve and how do you achieve it? And

And I look at it from you know, so then from from like enterprise network access and you're planning like how do you protect your data, right? Because ultimately it's like resources and data, you can use DLP, but then you can use CASB, but then you can use an SDP, but then you can put the entire thing through ZTNA, and then you can take and go into my y you know, it's like it how in the heck I know what I would do, but like how in the heck is a like

I say normie, but like a like a non full time security practitioner supposed to make heads or tails of all this stuff. Like are we shooting ourselves in the foot? Yeah.

Tony Lauro (01:20:50.135)
you know the the funny thing is, think of your latest and greatest data breach. Did any DLP stop that from happening?

Max Clark (01:21:00.45)
Well, first off you have to define what the heck DLP is. Right. So

Tony Lauro (01:21:04.078)
Where it's applied and all that. Yeah. So, but I even will go even furthermore to say, latest and greatest data breach. Did any antivirus, if that's even a thing anymore, did any malware prevention, did any EDR, did any like these tools, they have to be working correctly, first of all. and we all know years in the business.

You buy technology, you deploy it, you get it about sixty percent, and then the next priority comes on. Yeah. So kind of our our own some of the things that just are part of human behavior and budgeting and the functionality of IT and security having to wear so many hats. Right. Is

If if we could start over and start from scratch and make every environment into like a military grade secure environment, the company probably wouldn't function anymore.

Max Clark (01:22:03.318)
No, of course not, right? You you're always there's always competing like usability versus outcome, you know, pr security, right? It's like always this kind of game. The EDR, I mean antivirus EDR. I I I have an insurance industry trade group really actually was an insurance carrier.

released numbers recently and was talking about EDR specifically, which I found really interesting. Yeah. And it was percentage of ransomware encryption at you know, events that took place that had a leading EDR in place, right? And I won't expand on that, but you know, the and and the numbers the majority. And the only protection against that full encryption event was

the companies that had EDR in place paired with an MDR professional services organization that was tasked in properly watching the security tooling to make sure that it wasn't being actively exploited. And when I get into this conversation now and we talk about, you know, stats around dwell time, right? And it got really freaky because dwell times are pushed pushed all the way up almost to a year, right? And that like you think about that number and you're like, it's terrifying. And now dwell times are back down into days and you're like, that's terrifying.

And I and I try to tell people in council of your goal now is not so much, I mean, your goal is of course prevent as much of the low-hanging fruit nonsense as you possibly can in the first place, but your goal is just is shifted to like rapid detection and containment. Can you identify this thing before it it it's you know, unmanageable and you can't recover from it's unrecoverable or

Tony Lauro (01:23:31.766)
Yeah.

Max Clark (01:23:44.268)
Can you catch it fast enough so that way you can contain it and prevent it from being like, sorry, we can't pump gas anymore on the eastern seaboard, you know, kind of thing, right? w which is which is terrifying.

Okay, let me think about how I want to what I what I want to get into here. So many notes. I can be doing this for days. So let's see, there's a so I had a CEO describe, and it was specific we're talking about tech purchasing, but he he he described his his experiences for questions, right? Is this the right one? Am I paying the right price with the best terms? How long does it take to implement? How do I shorten the evaluation process?

And and it was really you know, it's like you you intuitively know this doing this, right? But it really stood out. And now those questions apply to security and AI as much as anything else, right? So if you're a buyer and you're trying to work through that framework and you're saying, you know, let's start with, is it the right one? Like how how in the how is a s CISO or a a a security buyer supposed to have

Like any idea of deciding is this the right AI security investment I should make and do I get a meaningful output from it? Like you know, how separate noise from like wheat from the chaff.

Tony Lauro (01:25:03.618)
Yeah. I think that the challenge there is that there are so many different things to take into account. Right. We we mentioned the role zero trust plays. I'll do hashtag and hashtag zero trust. Yeah, yeah. how your devices communicate to each other are absolutely a a a massive fundamental

Max Clark (01:25:18.542)
Yeah.

Tony Lauro (01:25:32.152)
visibility point you need to have, right?

When your LLM calls out to a model repository like huggingface.co or something, would it be interesting to know if your LLM ever called out to another one that was not authorized? You could easily tell that through a DNS request inspection, or through traffic inspection, identifying that that device is now talking to another system.

without any kind of system in place to automatically respond to those changes. you know, th again, we we talk about dwell time and about EDR versus MDR versus micro segmentation, even if we could put that in there. To some degree they all need to exist together because they serve a different force of function.

Micro segmentation, for instance, the the first thing we noticed after the acquisition, and we started to integrate and deploy this across our own environment, was we had this massive return, this discoverable, observable viewpoint of what actually was happening in the environment. Right. Not the application data flow diagrams, which I used to have to make and update, hey, the auditor's coming in in a month, and I'm like, my God.

Max Clark (01:26:50.892)
That physio's accurate.

Tony Lauro (01:26:52.854)
This is accurate, you know what I mean? Or the firewall rule set, which you know, I I I had a case years ago where we had a Cisco fifty five five go belly up, I had to replace the entire thing, and then I I go through to deploy the backup config and the backup config wouldn't work. It was the backup config for some reason had not been updated in god knows how long.

So I from scratch, and I I think it was six blades or something, I forget exactly how many ports, but I had to recreate this this whole thing. And I'm doing this on like a Saturday afternoon because we thought that we'd be done by Saturday morning because we started the the change window on Friday night. So I'm there all night. I'm there all afternoon, all day Saturday. and my boss is like, just make it work.

Max Clark (01:27:49.036)
Yeah, just make it work.

Tony Lauro (01:27:50.464)
And just make it work is code word in IT. Exactly. Yeah. Because if no one cares about IT until something doesn't work. I know. So this kind of force of function for like, you know, good, better, best becomes nothing at a certain point. Yep. so the observability that we got when we started to identify what devices are talking to each other, how they're devicing or how they're communicating.

Max Clark (01:27:55.431)
I don't care. Just turn it on so people can use it.

Tony Lauro (01:28:20.414)
using AI to label the applications that are talking, what version they're running based on information that they send, and then what users or what processes started those apps and what users started the process. If it's a system account or you know, would it be interesting to know that, you know, an SVC host that exe, which you see dozens of those running on a Windows server, that one of them was started from a DLL that was put in an attempt directory by a new admin that was just created.

Would that be interesting, right? But most most orgs don't have that level of visibility. You know what I mean?

Max Clark (01:28:57.1)
The problem with that trigger is trying to sell that. I mean, I mean, forget as like an outsider. Like if you're inside an organization and you're trying to sell that capability, you know, you're you're selling a

What what's the most common common vector in in security sales? Let's do a a a proof of value. Yeah. Right. We don't call POCs, we call POVs, right? And and what is the the the secret truth of every POV? I will find something because it's guaranteed it's there. Like you have some system talking to some country that you have no business talking to. It's guaranteed. And as the organization gets larger, the amount of it just increases, right? So every security vendor knows this and like wink, wink, nudge, nudge, like, let's do a POV and we'll we'll Yeah and then we'll we'll ge go through

Tony Lauro (01:29:41.774)
Almost like the joke's on you because you're gonna you're gonna see something, right?

Max Clark (01:29:46.348)
Yeah, but outside of that, when you go through a purchasing, right, and we get back into you know, budget justification, right? You know, either you've managed to find something that you're replacing, so therefore opens a budget for you to then spend because you're not going and out and asking the organization for additional money, or you have to justify it and you say, This is what we're gonna get out of this capability. And you're saying, you know, to to

you know, again, if you haven't had that experience of having some sort of breach and saying, we should be checking to see if, you know, there's there's you know, whatever like your scenario, like we have a new admin account launching a, you know, f fake SVCO, you know. How do we sell that to a non-technical buyer, let alone a non-technical buyer who hasn't been through a breach?

Tony Lauro (01:30:32.908)
Yeah. you know, I usually let the salespeople deal with the actual sell part, but the value part that I usually talk about is really, really specific around having and this is not like a you know

Max Clark (01:30:41.188)
Right, Sam.

Tony Lauro (01:30:56.802)
you know, a a device registration database or anything like that. This is actual observability of what's happening, automatic policy creation, and then automatic response to something that goes wrong. So if someone clicks on something, that malware, you don't have to have a team searching for it. That malware, what it's doing, gets stopped at a localized level.

on on the the breach of that one device. Right. the observability piece is usually the the wow moment where people are like, I didn't know that these systems were doing this or I didn't know that system was still functioning. You know what I mean?

Max Clark (01:31:41.654)
We saw this with asset and vulnerability management and log4j, right? Yeah. Companies that had asset and vulner you know deployed already, what what happened? Well, it turned out that they were running four different versions of Log4j on every single device in the organization and had no idea. Yeah. You know, it was like and it was it was amazing to watch the whack-a-mole of like we chain we we found one version, we t we we remediated. And the people without it were struggling to find it because it was such a big deal.

Tony Lauro (01:31:45.164)
Yeah.

Max Clark (01:32:07.552)
With you know, in tech, when you get into a category, you know, like we can talk about whether or not vendors are actually being truthful and what their product actually does, but

People enter a category, the category gets defined, you have competition, technology then becomes the great equalizer in terms of value and price gets driven down. When we're at the early parts of a buying cycle, AI security, how do we do protect our LLMs? How do we protect our users interacting with LLMs? How do we protect our brand from being a you know somebody abusing our LLM? We haven't hit that saturation point yet of having defined how how do you figure out like a price point.

Is this price appropriate for me to pay? Like, and that's not going to be necessarily against this company is selling it to me for this price and this company's selling it to me for this price, and my procurement team got me a 15% discount. And so therefore, yay, you know, like how do you but but at a higher level, how do you make that decision of this is worth investing in because I'm gonna prevent some something terrible from happening.

Tony Lauro (01:33:15.138)
Yeah, you know, w we we saw this in the early day of early days of WAF as well, right? Where I would talk to a customer and they're like, We have WAFs. This is where they sit. and then you kinda go through and you start doing some discovery and they're like, well, they're not they're not actually blocking traffic. Yeah.

Max Clark (01:33:39.288)
They're not doing anything. They're there.

Tony Lauro (01:33:41.698)
The early days of PCI, it was like Yeah, sorry. I don't mean to trigger you. But like, do you have a WAF? Was one of the first things. And PCI, the the first version was like, Do you have a WAF? Yes. The next one, the wording was different. Do you have a WAF and is it in line? Mm-hmm. It's like, Yes. Do you have a WAF? Is it in line and is it actually blocking traffic? Like you had to have some sort of enforcement check, right?

The the same thing tends to happen when you start to look at the big picture of enterprise security controls. and the real challenge to anybody, I used to have this presentation, it was all Princess Bride themed. you're familiar with the movie? Great movie. Classic. And one of the one of Wesley is being tortured in the in the dungeon in the woods there. And

Not Humperdink, but the other guy that's torturing him, I forget his name. no, it was when Humperdink comes in there. He's riding on the tank and and I had a word bubble that was like, I thought I thought the WAF config wouldn't block any good traffic. I'm sorry. And he's like, You ruined the sales or whatever. but that was very specific audience.

Max Clark (01:34:48.13)
Yeah, and he goes up to eleven, yeah.

Max Clark (01:35:07.586)
But that was the secret of WAF for a lot of people, right? It like, Well, how do you know if your WAF is is set correctly? your sales and marketing is yelling at you to turn it down, right? I mean, that became like we just keep dialing it up and then it was like, I got yelled at, okay, dial it back down a little bit. Right. We found the organizational pain tolerance for, you know, good versus bad tab traffic. I'm laughing so much at your PCI reference because we went through I went through an, you know, dec decades ago, went through a a PCI audit and

And this is funny because it was triggered because the people filling out the questionnaire were like, do we have this tool? And they're Yeah, we think we have this tool, like check the box. And then of course, like the answers made no sense. And so we had to go through a full audit. And the auditor was telling telling me that we needed to deploy firewalls in between the application server. You know, it was traditional stack, right? Firewall, load balancer, web server, application server database, you know, very web two point zero ideas. And the auditor wanted firewalls in between the web and application and application database tiers.

And I'm sitting across a table from this auditor and I said, for what? You know, like the application server has to use name and password for the database. Like the firewall is only gonna say allow, you know, yeah. What are we getting here? Except it's not performative. You know, and even today I walk through data centers and you and they stand out like a sore thumb. It's like you see the cage you see the cage in the data center where they have the cage over the top.

And you're like, you like instantaneously know what that or that cage is doing because you know what regulatory, you know, audit requirement like triggered the ceiling to be put on. And I'm like, okay, we had to go through like fifteen man traps to get into this floor space in the first place. Yeah. And and it would be

Tony Lauro (01:36:50.326)
Very mission impossible esque if that were to

Max Clark (01:36:55.23)
I I a billion years ago had infrastructure in data center and somebody decided to push the Epos. Right? You know, and by by requirement, they have to be accessible. Yeah. And what happens when you pull up both lids and you push both buttons at the same time inside the data center? Right. gosh. And a predictable outcome, right? They they function as it's you're like, okay, well, what's their security measure against that one? You know?

Tony Lauro (01:37:24.014)
They were allowed access, they went through all those security mechanisms, but they performed a function that was stupid.

Max Clark (01:37:32.248)
By yeah. Well in the cloud deletes a valid operation. I keep saying this one nonstop, you know, like it's yeah, you know, like we saw different category but similar question, right? You see this big push before it was called AI, you know, with the the contact center vendors of being able to say, Hey, we'll automate your phone trees and we'll we'll do deflections and we'll come in and we'll we'll help we'll help create like these voice agents for you.

But then you get into it and you look at the line item you're like, wait, why is a professional services line item like 10x the actual cost of this thing? And why are you, you know, and and what do you mean this project's gonna take nine months to roll out, right? You find out that the complexity is actually a big deal. When we when we talk about, you know, this like next generation like, you know, of WAF, you know, really like this AI WAF. WAF's not the right term. This AI firewall, I'll just use that, right?

What's that deployment curve going to be for an enterprise getting these things out there in the w I mean, this isn't like we're gonna turn it on. You know, DLP, we joked about that earlier. Like how do you detect and protect DLP, right? Yeah.

Tony Lauro (01:38:41.23)
so typically the way it works is you would put it in line but not blocking and you can run different prompts prompt tests against it. and there's kind of a battery of tests from the OWASP LLM top ten. You basically run these different types of potential something that would trigger a rule, right?

And then certainly you watch the normal traffic that's coming to the LLM. And would this have triggered a rule? And then you kind of go down and you and you do an evaluation. You're like, okay.

I and again, same thing for WAF. Every single time someone goes to this page, there's a cross-site scripting attack that rule that's triggered. It's like, well, we actually are including some script in this page response for whatever function. And it's like, okay, that's a false positive. So it's gonna be a similar kind of rollout where you're you're saying, okay, let's see what the usage patterns look like right now. Let's see how it's being used.

most abuse against against LLMs because it's it's kind of systematic in nature, you can see when someone's trying to reprompt or or re

when they're trying to affect the system prompt or get access to the system prompt. because you're kind of referencing you're referencing information that is pretty clear and deliberate. I think what we're going to see in the future are attacks that, for instance, I talked about the need to identify

Tony Lauro (01:40:30.06)
what resources your LM is using. You know, we've we've had for years an S bomb, a so software bill of materials. So I think now they're gonna start having L L bombs.

Where you can basically have to have like a whole system of different visibility tools in in place to validate all the things that your LLM is supposed to and what it does communicate with. No one's gonna block an outbound call to GitHub. This is why people started posting malicious code on GitHub and then referencing it in their malware. The malware would make a call out to GitHub to download a command and control tool. If I put a malicious model up on Hugging Face and trick your

LLM into loading that malicious model instead of the one it's supposed to load, now you have the potential that now the LLM could be instructed to do something totally different. And without any kind of intervention in between those those things, you you wouldn't know. So the attack may not come in like a, you know, a a real plain simple. One of the attacks we were mentioning earlier with the chatbot with Chipotle, what was kind of funny,

There's another attack called Lord of the Rings attack where you basically trick the L L into reciting the Lord of the Rings, all the text, which obviously is incredibly expensive. So

Tony Lauro (01:41:58.894)
So there's a training period.

Max Clark (01:42:02.67)
You know, like immediately I just go to this with like are we selling are we are we creating another category to then sell another tool that's gonna end up in, you know, an overworked IT security practitioner's desk that doesn't have right. That like doesn't have time already in their day job to like even think about this. And now it's like all of a sudden you're like, you have to now figure out how to like validate and train and inspect and deal with this.

Is there any chance for an enterprise to be able to deploy and manage this internally themselves? Or is this just a c a really a tool that has to be deployed and managed by a service provider that then has some sort of contractual fiduciary responsibility back to the enterprise to say, we're gonna do this for you?

Tony Lauro (01:42:51.82)
Yeah, I mean, to some degree, I have always thought that having a service provider model of experts is actually a better idea than you trying to become an expert on everything. again, I'll go back to WAF. You buy these WAF boxes and they, you know, the vendor helps you with the traditional deployment, then they're up and running.

Did that vendor ever come back and say, Well, what rules set are you guys on? And what's the fidelity of those rules? Are you actually blocking what you should be blocking, et cetera? No, they just come three years later and say, I'm gonna ready to sell you the upgraded boxes. Right. So from that hardware versus kind of the monthly recurring revenue model, a service provider has a lot more skin in the game. One

you know, especially if you're already a a customer of in this case Akamai, we've already helped you build and deploy your application environment. We understand how your business model works. the integrated account teams have relationships with you going back ten years. Now you're deploying a new technology. The the goal of that service provider is to be that trusted advisor to help carry out what you need to get done on your behalf.

So to some degree there I think there's going to be a learning curve where most orgs are running so fast building AI, but they don't know what to do about it from a governance or a security standpoint. maybe those two things will shift after a while where they're better at governance and securing those systems, and they need less and less help.

In in my mind, that whole service provider model starts off heavy engagement and then as they're trained up, it starts to lower. And that seems like a natural path.

Max Clark (01:44:50.734)
I mean the service provider customer relationship is comp it's complex. It's complex for both sides. It's complex for the service provider, though. I mean, we have the benefit of of dealing with so many companies across our, you know, portfolio and categories, right? It it's it becomes relatively easy to sniff out like y you know, it's like you can you can very quickly figure out like, are you actually real or not? You know, I've done this enough times now, right? But

But it's it's it's always surprising to me when you have like what what appears to be a a segment leading company and you start asking really basic questions and you very quickly get to a point you're like, there is actually nothing here underneath the surface that I would purchase if I was buying this. And but also within that, like we talk about like famous hacks, right? We you know, and and and like lateral movement, you know, target, right?

you know, HVAC contractor gets compromised. What value does an HVAC company have? nothing. Well, it turns out they have something that's really valuable, which is access into, you know, e X amount of POS terminals because they can get into it. The thing that's not discussed about with that breach at all is there was a service there was a SOC that was contracted that was flagging activity and network traffic that was signaling back to the corporate IT team.

That didn't wasn't doing anything with it. Right. So now you end up in a situation where it's like, y you know, this whole like, you know, no nobody wants to buy a tool and be locked out of it. And and at the same time, you don't want to like you know just turn over the keys to some third party and say, okay, you're you're responsible for taking care of this without discussing this with us in internally, because you you know, like who's the actual expert of your own environment? But if the service provider

isn't in a position of being like, if you see something, do something about it, and then you call me. Right? Like, how how y you know, like, what do you what's going on now?

Tony Lauro (01:46:47.916)
I think that's where the depth of of knowledge for the service provider really comes in. f first and foremost, everything we do starts with how do we identify what to do on your behalf. because when you have any kind of contractual relationship, so f for instance on the on the DDoS standpoint for prolexic, there's a massive, massive amount of time with establishing a run book

You know, what kind of operations are we gonna have? If there's a call tree for calling someone to inform you of a problem, what happens if that person's asleep? Who do I call next? Yeah. so but the same type of thing has to take place on that application exposure side. Whatever you're doing, enterprise security, whatever the case is, if you're managing a function on someone's behalf, there needs to be some kind of model, rule of engagement that shows

Max Clark (01:47:25.486)
You know, it never happens. Yeah.

Tony Lauro (01:47:45.282)
You know, first of all the tools should be doing things automatically. Right. Any tool, the best tools, in fact, sometimes, you know, in the past Akamai has people like, Well, they're kinda moving kinda slow and everything. I'm like, if you're a small company and you're trying to get, you know, famous and get bought out,

it's very advantageous to have just an army of people doing custom configurations. Right. We saw this in the bot game for quite some time where we had competitors that were kind of, you know, coming after customers, but they were doing manual everything. And that seems great until they get more than ten customers. Now that company has two thousand customers, what happens now? You can't do it. So building the tools to work on behalf of the user, on behalf of the the company themselves

To have an automated response that's intelligent, that's yeah, it's it's flagged based on these different categorizations, and you can see what those categorizations are and why it was blocked. I'll take that ten times, you know, nine times out of ten. because you don't want to get a call from a service provider. You want it to just do its job. Right. so part of it part of it's planning and then kind of the operational aspect of what do you do next, you know.

Max Clark (01:49:03.702)
I you talk about like percentage of tools that are not actually fully deployed, right? I mean, I I can't tell you the amount of times I've gotten a call from a CIO, a CISO, and and said, we bought this service and it never deployed and now we're trying to evaluate a replacement because our contract's up. But like their number one evaluation criteria is is this actually gonna deploy this time or not? You know? And and I th but I mean I feel like that's only part of the part of the thing. It's like if you're not talking with a service provider and you're going through implementation and if they're not being incredibly opinionated on how they're gonna support you.

You know, isn't that now like the responsibility of the service provider? Of like not to like treat your customers like their children, but at the same time be like, no, we will not do that and this is the way this has to be, otherwise this just won't be successful because it's bad for the service provider as well. Like if Akamai signs a contract with somebody that then doesn't implement what you want to implement in order to give them the value that you've sold them, you get revenue, but that customer will churn and be pissed off, right?

Tony Lauro (01:50:03.074)
Yeah, a lot of that has to do with the relationship you have with that customer. Right. I've been in IT since the late nineties. I've dealt with a gazillion vendors and I always thought it was funny when they come in, they're like, you know, they kind of tell you what you want to hear and all this good stuff. when I started Akamai thirteen years ago, it was the first time one I ever worked for a vendor. It was also the first time that I kind of saw the depth of

the bench, if you will, of the type of people, the consultative nature of those people. anyway, long story short.

When your service provider engages with you, you you have to have a trusted relationship. It can't be, we're coming in to tell you what the rule of law is and you have to follow it. Right.

You need to have trust, you need to have open communication, this is like marriage counseling, you need to you know what I mean. But part of that that consultative approach has to be a a very personal thing. Remember, service providers are coming in and in some minds of IT owners, they're displacing a function that they could be doing themselves.

and this was the whole thing for years why a lot of people didn't move to the cloud right away. I'm not putting my data out there or or I'm not gonna do this, like I can do it myself. And then they waste all these cycles building something when a corporation has already mastered it within six months, right? So

Max Clark (01:51:35.574)
It is it is actually it's it's become the I think the easiest signal for maturity of the buyer for me, of the build versus buy question.

Yeah. You know, and when you start talking about especially buying and implementing tooling in inside of IT, like like instantaneously you kind of get an idea of maturity of that organization. And when I say maturity, a lot of it's just they've been through enough pain, right? You know, like you can and and look, there's there's there's scales to this.

You know, if you've got a hundred person IT team, your capabilities of of executing program are very different from, you know if your staffing ratio is fifty to one, you know, your capability of deploying that is very different as well. But then you've got other up pressures because now your budget increases and you're and and then the organization is looking at that budget expense of like as a support function and like are you know, like there there's always this like, you know, like like fluid river movement that goes on in these things. But you know, we talk about like

Almost every category of security for me, MDR, you know, bot management, WAF, like all e you know, AI security now, like all these different things, SASI rollouts, you're like when you hit scale, because you've been able to operationalize your environment to hit scale as a service provider,

That's a really good thing that then drives cost down. And that I don't think people understand that or think about it right way. You know, you go to you know, like in order to hit scale and have a thousand customers, you've had to operationalize your process and build efficiency into that process, which means your cost to deliver that goes down, which then generates margin for you, right? But like, you know, like th these for me it's these are all become really good things.

Max Clark (01:53:23.054)
We've both been doing this for a long time. And my joke with IT becomes like what's old is new again. It's like the new hotness was the old hotness, right? So you've been doing this for a a long time now. DDoS, WAF, bot management, API security, microsegmentation. Now we're talking about AI, browser-based plugin security. Like

Tony Lauro (01:53:48.234)
It

Max Clark (01:53:50.176)
Is this just the same problem being exposed or being s you know, it's like is what's old new again and now we're just attacking it to a different layer? yeah, yeah.

Tony Lauro (01:54:01.016)
Need some water. Do you have something by chance? Sorry. Yeah, if you don't mind. Yeah. I got one right here. I don't know if it's something with the AC. Maybe that's it too. Thank you, I'm sorry.

Tony Lauro (01:54:26.422)
That's why they put tape over all the logos whenever there's

Max Clark (01:54:29.046)
Yes, because two ac two things, either they've got an existing endorsement deals or because it'll get hit for brand infringement and decent

Tony Lauro (01:54:40.302)
So even the logo on the back of like Apple computers, they cover it up.

Max Clark (01:54:45.304)
Some some places are way more aggressive with it when you start talking about like what sort of enforcement actions you have.

Tony Lauro (01:54:51.5)
And and sorry. No. What they're what they're running against. Yeah, yeah.

Max Clark (01:54:58.895)
I usually had water like laying around that was unlabeled, but I apparently neglected to pull that out. Yeah.

Tony Lauro (01:55:07.022)
Yeah, I don't know why. Just kinda like a

Max Clark (01:55:09.316)
there's something in the air. I had a day the other day where I was just like, I just I I just buy the industrial strength sudafed at this point. Yeah. You the tw load up. The 12 hour stuff is amazing. Yeah. Like you I I'll have a thing where it's like I'm not feeling 100%. It's like, have I had enough water? Yeah. Did I eat well? Did I not exercise this week? What's going on? And then you're like, I don't know. And then you take a Sudafed and you're like, I love I love everybody. You know, like I love everybody.

Tony Lauro (01:55:15.918)
Yeah.

Max Clark (01:55:35.448)
I need to remind actually I need to tell my wife this if I'm particularly cranky, you know, it's like just Yeah, you know, I don't have a I don't have bad allergies, but there's something that hits that just kicks in and you're like

Tony Lauro (01:55:39.286)
Yeah.

Tony Lauro (01:55:46.85)
weird yeah I'm trying to figure out what it was it's just that there all right what was the last question you asked

Max Clark (01:55:50.05)
Yeah.

Max Clark (01:55:53.834)
I you know, are we just dealing I mean like is this is is time a flat circle? You know, are we are we really are we are we really pushing into something new or is it just like old is old is new again and we're just reinventing and solving the same problem over and over.

Tony Lauro (01:56:08.898)
Yeah, I think that you know, I mentioned this earlier with kind of the the deployment cycle of new technology meeting butting heads with the old problem, the human problem, which is we don't trust the technology or we've got too many you know, too many objectives that we have to meet and not enough budget or time or people.

So to some degree, we're facing the same problems again, but we have the opportunity not to make the same mistake. and and time has repeatedly told us that we will make the same mistake over and over again. So it's a tricky response, you know.

Max Clark (01:56:57.102)
More I do this, the more empathetic and sympathetic I am. It's just, it's impossible. You have an impossible problem. You know, you talk about like if you're if you're serving an application, you talk about like your banking application, you now have you now have to support a very broad audience of customer with way disparate education levels and and capabilities, right?

And at the same time protect them in a way that doesn't piss them off at you as best as you possibly can. And like then figure out how do you sit in the middle of that. And you know, it it's it it's really it it's just, you know, it's tough. It's really tough. let me let me final question, let me wrap with one thing. I think we've

Tony Lauro (01:57:48.93)
to to that what you just said, a funny quote that sometimes my wife is like incredibly intelligent and sometimes I'm just not getting what she's saying. And she'll say, I can explain it to you, but I can't understand it for you And I'm like, I I think I'm offended actually. No, but it's really interesting because there are times where that's that's where you really have to be

you really have to come in with the right attitude. Because some of the things that we're doing and the changes that are happening are happening so quickly. Not everyone's out there always talking with you know,

learning groups and going to conferences, they don't have the same resources and access. So you really have to come to that to that problem as a consultant and as a a friend even to broker that conversation to actually help it make sense to them because there are a lot of complicated things going on and it's moving so fast people are kind of overwhelmed. So I think we have to be kind of gentle, I guess, with that approach, if you will.

Max Clark (01:59:01.162)
I I've okay, let me think I mean let me let me craft this make this intelligent question here. For a long time, the the there was a push in security that was around this idea of user education, security awareness training. We can educate our user into a secure environment. That is the biggest crock like that we've perpetuated on and you know, you know, like the it and it's it but it's

N it's a requirement, right? Security awareness training is a requirement for om for every cyber insurance program, right? Like it it's it's it's become like

I, you know, and San I think it was Sans that published this a billion years ago, was talking about like threat management, what the threats the organization were, and you and and the and the and the reality is is the threat vector hasn't changed in decades, right? Like it's always users. Users are the soft spot, right? Whether it's intentional, accidental, malicious, you know, like whatever it is, yeah, you know, it's it's like you know, it it always comes back to the same place and is the

Does the does this need to shift? You talk about like education, right, and and and helping people, right? But you're not talking about helping an end user inside the organization. You're talking about helping the IT the technical teams in educating them into what they can actually do. And it really feels like the the the the correct goal here is educating the IT practitioners and how to protect their users almost without the users aware of what's actually going on.

And staying in this level. I mean, I think you used earlier, it's like almost invisible. Like how do you you know, so is that the real goal? Is the real goal here for tech to stay completely invisible inside of the organization and just do things that users aren't even aware of as occurring and then, you know, taking that approach instead?

Tony Lauro (02:00:59.17)
Yeah, I mean I think we could take lessons from kind of how people are starting to like the latest and greatest technology is AI, right? And when you ask something of an agent and it gives you a response, you don't care to understand how that's functioning. Well. I mean, we might. In fact, I want to know w way too much about how that's functioning. But so the idea of

your tools working on behalf of the user. The idea there is that the technology should be kind of seamless, but the customization should be available. and that's kind of something I think that over the years, when I first started at Akamai, we had just deployed WAF.

And we were just thinking about getting into bot management. So there was two security tools, functions, if you will. and over time we, you know, rotated one direction where we're like, you can do absolutely anything. Here's how, you know, customization, this and that. we over-rotated that way because we knew that people wanted that deep level of of customization because every business is different, every web environment's different.

And then we kind of slowly kind of rotated the other direction where we're like, we have now been doing this for, you know, almost fifteen years. We understand the intent of these different bots based on tells and signatures and and timelines, etc. so that

you can use the collective intelligence of everything that we've learned so far and just apply what do you want to get out of this? What is your end goal? And then have the tools do the work on your behalf. Because if if everything is almost like if there's too many options, there's no options to some degree. Right. and from from that perspective, making the tools work

Tony Lauro (02:03:17.45)
a little bit more autonomously and I know that sounds scary when we look at, you know, automatic, like using AI and ML to discover

new attack techniques and then apply rules automatically, you should probably still have the ability to say, here's a rule that's about to be applied automatically. Please review it and then let it, you know, get deployed. There might be a human in the loop kind of situation where you say, that's actually really intelligent. This was a new remote file inclusion attack I hadn't seen before because it's never been, there's no C V E attached to this particular exploit type, right?

Y you should probably still have the ability to review things. and we certainly we do things automatically and we can push things automatically.

But having the users just the human nature of you being able to to make a decision and say yes or no or to have insight so you can have a proper evaluation, I think that's the difference between, hey, trust me, you know, black box security, trust me, we'll take care of it, you know. Yeah. you know, salute. Or or let's work together. And let's work together might not mean you have to do a whole ton or we have to do a whole ton, but let's communicate on why decisions are being made.

Max Clark (02:04:38.69)
What what I hear from you more than anything else becomes like it's like the defaults matter. And I don't and and they actual and how and and just how important the default is is

probably the most important thing actually in IT and and and technology point of it right so the the default is the most important thing and then having an opinionated default or an opinionated push you know from an from an Akamai is really important but then giving flexibility to the customer to then to selectively opt out or customize the default but it it but it it's like it shouldn't be the other way. It shouldn't be like you know here's this Swiss Army knife now config

Tony Lauro (02:04:56.942)
Made a

Tony Lauro (02:05:19.586)
You can do anything. You just tell me what to do. It's like why did we hire you? Yeah.

Max Clark (02:05:23.596)
Nobody knows what to do, right? Tony, thank you very much. Of course. That's it for this episode of Signed. If you got something out of this, share it with someone in your world who's staring down a tech decision. A CIO, a CFO, a founder, a procurement lead, whoever. That's how the show grows.

Tony Lauro (02:05:29.304)
Really appreciate it.

Max Clark (02:05:42.09)
Everything from today lives at itbroker.com slash podcast. Show notes, transcript, links to anything we mentioned. If you're in the middle of a real tech decision right now and you want someone in your corner without the vendor bias, that's what we do at itbroker.com. Schedule a call on our website, buy tech without regret. I'm Max Clark. Thanks for listening. See you on the next one.

Tony Lauro (02:06:01.558)
Yeah.