David Bombal interviews Radware North America lead Randy Wood on what AI is actually doing inside enterprises, framed as a 1994 internet moment times a thousand. Wood walks through the five concerns customers voice every week, data, identity, visibility, the threat landscape, and the loss of autonomy, and explains why shadow AI is the new shadow IT. He argues guardrails fail against autonomous agents and Radware instead validates behavioral intent before execution, then ranks DDoS, API, and bot attacks as the top vectors and calls API security a board mandate, since firms that think they run a thousand APIs often have fifteen thousand. He closes with three predictions and advice to young people to make peace with AI without outsourcing their thinking.
Published Jun 14, 202631:06 video25 min readAdded Jun 16, 2026Open on YouTube →
At a glance
David Bombal sits down with Randy Wood, who runs Radware's North American business, for a wide ranging field report on what AI is actually doing inside enterprises right now, not on YouTube debates but in the rooms where Wood meets Customers across the US and Canada. Wood's framing line is the spine of the whole conversation: this is "a 1994 internet moment times a thousand," a time when people are curious, optimistic, hopeful, and terrified all at once, and they genuinely do not know what they are looking at.
The interview moves from that emotional weather report into five concrete concerns Wood hears from Customers, the same five every week: data, identity, visibility, the threat landscape, and the loss of autonomy. Inside those five sits the title idea. Just as "shadow IT" once meant tools running in your environment that nobody sanctioned, shadow AI now means AI tools and agents already loose inside the company that nobody mapped, secured, or even knows exist. From there the conversation widens into how attackers are weaponizing AI (phishing in flawless Finnish, smarter DDoS, the looming agentic attack), why API security is the underbelly nobody inventoried, why Radware bets on validating an agent's behavioral intent instead of trusting guardrails, and what Wood tells worried young people who watched their classmates boo AI off a graduation stage.
This page rebuilds the conversation in the order it happens, attributing each point to Wood or Bombal, so a network engineer gets the full Shadow AI briefing without watching first. The frame is a practitioner talking shop about a fast moving problem, optimism and dread held in the same hand.
Figure 1. The five concerns Wood hears in every Customer conversation, all converging on the shadow AI problem: AI tools and agents already running inside the enterprise that nobody inventoried. Each concern is a different face of the same blind spot.
The 1994 moment: curious, hopeful, and terrified
David Bombal welcomes Randy Wood, who introduces himself as responsible for Radware's North American business, covering enterprise, service provider, and increasingly mid market Customers across the US and Canada. Bombal sets the assignment plainly: forget the YouTube debates, tell us what is actually happening out there with AI.
Wood's answer becomes the title card for the whole interview. It feels, he says, like a 1994 internet moment times a thousand. People are curious, optimistic, hopeful, and terrified at the same time. They do not really know what they are looking at, they do not know what is about to happen, and they do not know how to orient themselves to this thing called AI.
Two data points crystallize it for him. His daughter is a high school history teacher, and through her he learned that 85 percent of all high school students use AI to write some form of paper. Held against that is an article headline he read, which he grants is hyperbolic: "The substitution of LLM for genuine thinking is the biggest crisis of our time." At risk, the piece argued, is the life of the mind, a generational crisis. Hold those two facts in the same space, he says, 85 percent of kids leaning on generative AI, and a chorus warning that we are learning how not to think, and you have the two extremes of where AI sits right now. And that is only the generative side. Agentic AI is coming on strong and consuming all the oxygen, with everyone trying to figure out how to deal with it.
Bombal agrees the outsourcing of the mind worries him too. Wood frames the corporate version: ignoring AI is heresy inside organizations now, because "figure out AI" has become a board mandate. People do not even know what compliance means. Is turning on Copilot enough to call yourself AI enabled, or is something bigger required? On his walk over to the interview, he passed three presentations, all of them about AI, all opening the same way: everybody is talking about AI, and here is my AI point of view on whatever technology I happen to sell.
Five concerns Customers actually voice
Bombal pulls it toward Radware's lane: from a cyber security company's point of view, this has to be a worry. A huge worry, Wood says, and he boils what he hears from Customers down to five recurring concerns. Radware, he notes, is increasingly in the AI and the API security business as much as classic cyber security, because the two are tightly linked.
One, data. The single biggest Customer concern. Securing it, and keeping long term control of it, because data is the lifeblood of AI. Wood's example is personal: his son is a young naval officer on the special operations side, and the unit's fear about AI is not what it can do for the mission but the risk of someone accidentally pasting confidential information into an AI prompt. Once that happens, the problems multiply.
Two, identity, and increasingly the identity of the agents, because, in the line Wood returns to twice, sooner rather than later agent equals user. Everything you do today to establish and validate a human identity now has to extend into the agentic world: access control, validation, the lot. The industry has not figured this out.
Three, visibility. You cannot secure what you cannot see. How many agents are running in my environment? How many APIs? AI is enabled and made possible by a sophisticated world of APIs, so the two questions are joined. And this is where the title lands: you have heard of shadow IT, so now we are dealing with shadow AI. How many AI tools are out there that I do not even know about?
Four, the threat landscape, which has changed dramatically. Wood's favorite illustration: for the longest time the least phished country on earth was Finland, almost zero phishing attacks, because the way you spot a phishing email is the mistakes in the language, the clumsy translation. Two years later that is gone. Finland is right up there with everyone else, because AI leveled the playing field. AI does not make mistakes in Finnish born phishing attacks.
Five, the loss of autonomy. The man in the middle, the human element, is disappearing. These agents operate and reason autonomously, which flips the whole security model on its head. Taken together, Wood says, the five demand a completely new approach to cyber security, and that is the niche Radware has carved out.
Are companies really deploying agents?
Bombal asks whether companies are actually deploying agents. Yes, Wood says, though, "truth in lending," in some cases it is a solution looking for a problem. Companies are still working out what they would even do with agents, and he is not sure the consumer use case will not end up more prevalent than the enterprise one.
He draws the key distinction cleanly. Generative AI outsources the ability to think. Agentic AI outsources labor. That distinction is why he does not believe AI will replace people at the scale predicted. There will be some reduction and consolidation, but the durable use is agentic AI augmenting how we think and how we do our jobs, not erasing the jobs.
AI attacking AI, and a teacher's lesson plan
Bombal asks if companies are being attacked more through AI. One hundred percent, Wood says, and points back to the DDoS world. Those attacks are getting more sophisticated because of the innovation in AI: not just high volume, but innovative, creative attacks. The mission becomes staying up, keeping pace, and that means fighting AI with AI. When you think you have one attack figured out, you have to reorient to the next, which arrives in a different shape. It is literally AI squaring off with itself.
Asked for a concrete story, Wood admits we have not yet seen agentic AI's real benefit at scale, then gives a small one that he loves. A high school criminal justice instructor, a 28 year retired sheriff's department veteran, went into ChatGPT and wrote careful prompts asking the model to build a one hour lesson plan on the proper use of handcuffs, broken into segments, 10 minutes of this, 5 of that, 20 of the other. It produced a remarkable plan, as if it had been teaching forever, and a guy who genuinely knows his business put it to work that same day. Then he ran the identical prompt for a class on administering a lie detector and got an equally good plan. A great use of AI.
But Wood balances it immediately with the education downside, because it is simple and everyone understands it. With 85 percent of kids using AI, the threat is to the mind itself. His daughter's read: kids want instant gratification, and in thinking, reasoning, and writing, the learning lies in the struggle. AI removes the struggle. The old saying goes "I know what's in my mind, I just can't write it down," but some argue that if you cannot write it down, you do not actually know it, because the act of composition is the thinking. Kids going straight to AI skip the reasoning, the thinking through, even the philosophy. He cites research out of Georgetown finding that a human written essay contains eight times more relevant, fresh ideas than an AI generated paper. Teachers, he adds, can increasingly tell which is which.
Metric
The number
What Wood reads into it
High schoolers using AI
85% write some paper with it
A generational risk to the life of the mind
Human vs AI essay quality
Human has 8x more fresh ideas (Georgetown)
The struggle is where the learning lives
AI favorability (NBC poll, n=1000)
26% favorable, 46% unfavorable
Polled worse than Trump, ICE, and Harris
Mention AI in an earnings report
Stock up ~25%
The hype is real and inflated
Figure 2. The four numbers Wood uses to map the public mood: heavy adoption, a real quality gap, deep unpopularity, and frothy markets. His verdict is that the truth lands "somewhere in the middle."
The public mood: AI polls worse than almost everything
To show how the public feels, Wood cites a recent NBC News poll of a thousand people on AI favorability: 26 percent favorable, 46 percent unfavorable. The real aha, he says, is the comparison. AI polled worse than Donald Trump, ICE, and Kamala Harris, and only better than the Democratic party and Iran. For the US, that is striking.
He points to the college graduation videos where speakers got booed at the mere mention of AI, in one case "the architects of artificial intelligence" announcing an AI reader system, booed mid sentence. The graduates spent four years and a lot of money studying something and now fear AI will take the very job they trained for. Wood expects the truth to settle somewhere in the middle. Right now it is hyperbolic, awash in investment, where mentioning AI in an earnings report bumps your stock 25 percent. But it is becoming an agentic world, and that is where the innovation will pour. Which raises the real question: how do you secure it?
Protect and serve: why guardrails are not enough
Here Wood lays out Radware's thesis, and it is the technical heart of the interview. Securing an autonomous agent is a completely different problem, because the agent reasons on its own, takes its own action, and has access to your confidential data. Back to the lifeblood, back to data. The stakes are very high.
The world tends to approach agentic security through guardrails, rules you create. Radware found that insufficient. As Wood puts it bluntly, the agents do not pay attention to the rules, they do not pay attention to the guardrails, it is not efficient. So Radware's approach is, in his word, radically different, drawn from years of fighting volumetric DDoS, doing bot mitigation, and running a web application firewall. They come at it from an intent perspective: validate the intent of the agent, behaviorally, before execution. The algorithms judge whether the agent is doing what it was designed to do, rather than trusting a rule to hold.
He splits the strategy into two postures, protect AI and serve AI. Protecting agentic AI is the behavioral validation just described, and Wood claims a head start as enterprises wake up to the problem. He stresses it is genuinely hard, complicated work, and different even from bot management, because bot mitigation is prompt based while an autonomous agent has to be met on its own terms, behaviorally. Serving AI is the other half: Radware wants to enable the agentic AI economy, not impede it, not strangle it in guardrails and rules. So the serve side brings the full portfolio, DDoS mitigation, bot mitigation, web application firewalls, to keep the agentic experience flowing without obstruction. Protect and serve.
Figure 3. Radware's bet. Guardrails put rules in front of an agent that ignores them; the alternative validates the agent's intent before execution, behaviorally. The wider strategy splits into protect AI (behavioral validation) and serve AI (the existing portfolio keeping the agentic economy unobstructed).
What is hindering deployment: no policy, no architecture
Customers are looking hard at agents but not deploying much yet, Wood says. What hinders them is the absence of formal policy and the architecture underneath it. The underpinnings have not been thought through, because there is a rush to market with everybody talking AI before anybody designed the foundation.
His analogy is geographic and historical. He lives in Northern Virginia near Loudoun County and Ashburn, where all you see are data centers, now spilling into Prince William County. The infrastructure is being built out everywhere, which reminds him of the dark fiber era, when companies laid fiber simply because they felt they had to have it. He expects AI to right size itself the same way. We have the Lego blocks now, and it is time to start building, but the policies are missing, and many companies do not even know which business problems they are trying to solve.
Cloud Mythos and the great awakening
Bombal asks about Cloud Mythos, and Wood takes it as a vivid example of AI cutting both ways. Alliances are forming, he points to Glasswing and says it is up to 16 or 20 companies now, including Cisco, that have agreed to use AI to audit and fix their own code, code they had believed was error free and bug free. AI has changed that assessment dramatically, so they are collectively committing to improve code quality and security. That is AI doing a very good thing while everyone simultaneously recognizes its potential to be used for harm. Cloud Mythos, Wood believes, is and will be game changing in a positive way, but if that kind of capability gets into the wrong hands, it is a nightmare doomsday scenario.
This is the "great awakening," he says. The cyber security imperative for AI is a whisper right now, but it goes from whisper to scream very quickly. We have not yet seen one of the big AI attacks, the kind that has hit utilities and banking on the classic cyber side over recent years. When that lands on the AI side, the industry doubles down. Cyber security always moves whisper to scream, and Wood feels that scream is imminently ahead of us.
What attackers are doing with AI today
Bombal asks the direct question: phishing, malware, what are you seeing? All of it, Wood says. Phishing, malware, and notably better phishing attacks in Finnish, his callback to the Finland point. AI makes attacks more sophisticated and lets them be launched at unbelievable scale. And in the hacker world these techniques are freely shared, so the know how and competency advance very quickly across the whole attacker population.
On DDoS, Wood admits he once thought of it as almost a poor man's attack, but it has become far more than that. It is increasingly sophisticated and innovative through AI, and the volumes are enormous. He frames it memorably: DDoS is simple, not easy. Simple to launch, not particularly easy to defend. This is Radware's core competency, hybrid DDoS mitigation, on premises hardware that can escalate to the cloud, and he claims they do it better than anybody. DDoS mitigation is here to stay, and AI is only making the attacks more sophisticated.
Pressed on what "sophisticated" means, layer 7, unique unrecognizable patterns, Wood says everything in between. The traditional layer 7 DDoS attacks the industry has fought for years now carry an undetectability that comes from AI advancing the capabilities dramatically. Radware's edge, again, is the behavioral approach: the same method that made them successful in DDoS, carried into AI security, and equally effective against the attacks of today and tomorrow.
APIs: the biggest threat nobody inventoried
Bombal turns to APIs, and Wood does not hedge: in present time, the biggest threat enterprises face is API security, period. The relationship between API and AI security is very tight, which is why Radware insists on solving both rather than one.
The root problem is how APIs were built. By and large they were not written with security in mind. They were written quickly and hastily to solve a function, and they did. What gets attacked is the business logic of the API. The discovery gap is the part that lands hardest. When Radware asks an organization how many APIs it has, the answer is often "maybe a thousand." Run a mature discovery tool and you find 15,000. Then you show them the vulnerabilities living in those APIs, and it is breathtaking. The elegance of these attacks, Wood says, is in their simplicity, vulnerabilities so basic you would not believe they exist at that level until an inventory, discovery, and vulnerability assessment opens your eyes. His prescription is blunt: API security should be a board mandate in every company right now, before you even think about your agentic AI strategy. Get your API house in order.
Figure 4. The visibility gap Wood calls breathtaking. Ask an organization how many APIs it runs and the answer is around a thousand; a mature discovery tool finds fifteen thousand. Every undiscovered API is unmonitored attack surface, the API face of the shadow AI problem.
The top three attack vectors
Bombal asks for the top three most attacked vectors. Wood ranks them from Radware's vantage. DDoS is first, present in every Customer they protect, and the volume they see and prevent is enormous. API is a very close second and may well eclipse DDoS, simply because of how much poorly written, vulnerable API is out there. Third is bots and bot mitigation, which is hard and manpower intensive because you must always stay one step ahead of the attacker, and it is heavily data science driven. DDoS, API, and bots are the three he sees not infrequently.
He underlines what he sees as Radware's difference: they are in the fight with Customers every day, meaningfully fighting DDoS, API attacks, and massive bot problems, and very soon the agentic AI security attack that is right around the corner.
The jungle drums: three predictions
Bombal asks what is coming. Wood, channeling the jungle drums from Gilligan's Island that warned when bad things approached, gives three predictions.
One, the AI bubble bursts, in a good way. Not a Y2K scale problem, but a right sizing. After it, the real work begins, and he expects data center consolidation (there is a lot of data center out there) and smarter processor capacity, with more innovation to come there.
Two, agentic AI security becomes a board mandate for all companies, because agents will find their way into our businesses and our personal and professional lives, and securing them will be one of the biggest challenges we face.
Three, agent becomes synonymous with user. An agent is a user. It has an identity, we give it access control, we treat it like a user, and we accept the loss of autonomy that comes with it. That is the trade Wood thinks the world is willing to make: surrender some autonomy for the simplicity and power of agentic AI.
Advice to the young: make your peace, keep your mind
Bombal raises the students booing AI, the sense among young people that their futures are being narrowed by what their elders built. He asks Wood's advice: is there a future in cyber security, in this industry, for young people?
A big future, Wood says. AI will become a meaningfully important part of our lives, and if you do not come to terms and make your peace with it, you will get left behind. But he holds the line from earlier: do not outsource your thinking. Generative AI will start recycling and generating slop if that is all we feed it. The good use is the opposite of laziness. His example: if you wanted to interview someone from history who is no longer alive, the genius is in writing five good interview questions yourself, then letting AI answer in that person's voice. That is a great way to learn and think. Outsourcing labor will find a meaningful place in how we work, but it will not eliminate jobs at the predicted scale.
He grants the young their right to be concerned, with a note of irony, that two days ago they were probably using AI to write their term paper. The graduation speakers who pushed back with a little tough love had it right: you are going to be okay, you are going to have to come to terms with it, and there are plenty of jobs AI will not replace. Even companies announcing a 10 percent reduction in force "because of AI" are, in his read, often blaming AI for a convenient narrative. AI is here to stay, it will right size itself, and over time it will make us all better at what we do.
Where to learn more
Asked where a viewer can go to learn about API security, DDoS, and the rest, Wood points to radware.com, organized by technology, with real technical depth across DDoS, API security, and Radware's new agentic AI security, plus occasional online presentations. Bombal thanks him and closes the conversation.
Key takeaways
We are in a "1994 internet moment times a thousand," where people are curious, hopeful, and terrified at once, and the AI conversation splits between generative AI (which outsources thinking) and agentic AI (which outsources labor).
Enterprises voice five recurring concerns: data (the lifeblood of AI, easily leaked into a prompt), identity (where agent equals user), visibility (you cannot secure what you cannot see), the changed threat landscape, and the loss of autonomy as agents act without a human in the middle.
Shadow AI is the new shadow IT: AI tools and agents already running inside the company that nobody has mapped or secured.
AI has erased the old phishing tells. Finland, once nearly phishing free because attackers botched the language, is now attacked as much as anyone, because AI does not make mistakes in Finnish.
Radware bets against guardrails, arguing agents ignore rules, and instead validates an agent's behavioral intent before execution, under a "protect AI and serve AI" strategy.
What hinders agent deployment is missing policy and architecture, not capability; Wood likens the data center build out to the dark fiber era and expects a right sizing.
API security is the biggest present threat. Organizations that think they run a thousand APIs often have fifteen thousand, riddled with simple business logic vulnerabilities, which is why Wood says it should be a board mandate before any AI strategy.
The top three attack vectors are DDoS, API, and bots, with API potentially eclipsing DDoS, and DDoS is "simple, not easy," sophisticated through AI and launched at huge scale.
Wood's three predictions: the AI bubble right sizes in a good way, agentic AI security becomes a universal board mandate, and agent becomes synonymous with user, identity, access control, and all.
His advice to anxious young people: make your peace with AI but do not outsource your thinking, because the learning lies in the struggle and human work still has eight times the fresh ideas of AI slop.
Chapters
0:00 Coming up
0:31 Intro
1:21 The reality of AI today
3:50 Customer's concerns about AI
7:12 Deployment of AI agents in enterprise
8:08 AI used to attack enterprise
9:01 AI as a problem and a solution
13:43 Agentic security with Radware
16:51 Lack of architecture and policy
18:06 The impact of Cloud Mythos
20:03 How do attackers use AI?
22:25 API vulnerability and security
24:38 Most common attacks
26:13 The future of AI
28:00 Advice for the youth
30:33 Where to learn more
Notable quotes
"I feel like this is a 1994 internet moment times a thousand. People are curious, they're optimistic, they're hopeful, and they're terrified. They don't really know what they're looking at." Randy Wood, 0:00
"85 percent of all high school students use AI to write some form of paper." Randy Wood, 1:50
"The substitution of LLM for genuine thinking is the biggest crisis of our time. At risk is the life of the mind." Randy Wood, reading the headline, 2:10
"The difference between generative AI and agentic AI is outsourcing the ability to think versus outsourcing labor." Randy Wood, 7:30
"Sooner rather than later, agent equals user." Randy Wood, 4:30
"You can't secure what you can't see. You've heard shadow IT. So now we're dealing with shadow AI." Randy Wood, 6:00
"Finland is right up there with everybody else because AI has leveled the game. It doesn't make mistakes in Finnish born phishing attacks." Randy Wood, 6:40
"The agents don't pay attention to the rules. They don't pay attention to the guardrails. It's not efficient." Randy Wood on why guardrails fail, 13:50
"We check and validate intent before execution. It's behavioral based." Randy Wood on Radware's approach, 14:30
"AI polled worse than Donald Trump, ICE, and Kamala Harris, and better than the Democratic party and Iran." Randy Wood on the NBC poll, 11:00
"You find out they've got 15,000 APIs. And then you start to show the vulnerability that exists. It is breathtaking." Randy Wood, 22:40
"It should be a board mandate in every company right now. Just API security, before you even think about your agentic AI strategy." Randy Wood, 23:40
"The learning lies in the struggle. And with AI, there's much less struggle." Randy Wood, 9:40
"If you don't come to terms and make your peace with AI, you're going to get left behind." Randy Wood, 28:20
Data Center Alley (Loudoun County and Ashburn), the data center build out Wood compares to the dark fiber era
Where it stands
This is a vendor conversation. Randy Wood runs Radware's North American business, and the interview naturally steers toward the threats Radware sells protection against, DDoS, API security, bot mitigation, and agentic AI security, and toward Radware's behavioral intent approach as the answer. That is worth knowing, and it does not make the observations wrong.
The directional claims hold up against independent reporting. The collapse of the old "spot the bad grammar" phishing defense as AI writes fluent text in any language is widely documented, the Finland framing is a vivid stand in for a real shift. The shadow AI and shadow IT parallel is a mainstream security concern, and the API discovery gap, organizations badly undercounting their own APIs, is a finding many security vendors and independent researchers report, even if the exact "1,000 versus 15,000" figures are Wood's anecdote rather than a published study. The "agent equals user" identity problem and the limits of guardrails are active, real debates in agentic security right now. The specific statistics he cites, the 85 percent student figure, the Georgetown "eight times" number, and the NBC favorability poll, are quoted from memory in conversation, so treat them as directionally reported rather than precisely sourced. Read this as a sharp, experienced practitioner's field briefing on where enterprise AI security actually stands, with the standard caveat that the practitioner sells the cure.
Full transcript
I feel like this is a 1994 internet moment times a thousand. People are curious, they're optimistic, they're hopeful, and they're terrified. They don't really know what they're looking at. Have you got stories or examples without mentioning names where AI has caused huge problems for companies or where AI has solved a huge problem for a company? The agents don't pay attention to the rules. They don't pay attention to the guardrails. It's not efficient. And so the Radware approach has been, we think, radically different. Again, it comes from years of innovation.
Everyone, it's David Bombal back with a very special guest. Randy, welcome to the show. David, thank you. It is just a real pleasure and an honor to be here with you. I appreciate it. I want to pick your brain about AI. So perhaps you can start like a 10,000 foot view and then we can get into the nitty gritty and specifically talk about cyber security and AI. But what I really want to find out is what's actually happening out there. I mean on YouTube we can have all these debates but you work with customers around the US I believe. So perhaps you can just tell us a bit about yourself and what you do and then we can get into the discussion about AI.
Yeah. Right on. So, Randy Wood, out of Northern Virginia, Washington DC. Glad to be here with you today. I'm responsible for Radware's North American business. So that's all customers in North America, US and Canada. That's on the enterprise side, service provider side, and increasingly on the mid market side.
So what are you seeing out there, what's actually happening out there with regards to AI?
Here's what I think. This is what I feel like we're looking at. I feel like this is a 1994 [internet moment]. People are curious. They're optimistic. They're hopeful and they're terrified. They don't really know what they're looking at. They don't know what's about to happen. And they don't know how to orient themselves to this thing called AI. And two things stand out for me. I talked to a lot of people. I have a daughter who's a high school history teacher where AI is a big thing, some for good and some not. And I learned that 85% of all high school students use AI. Wow. To write some form of paper.
Now, I'll talk about maybe why that's good and why that's bad. And I recently read an article. The headline was this. It was a little hyperbolic. It said, "The substitution of LLM for genuine thinking is the biggest crisis of our time." Yeah, at risk is the life of the mind. A generational crisis. So you hold those two things in the same space, 85% of kids using generative AI and you've got lots of people saying we're learning how not to think. We're outsourcing our thinking. And so that's sort of the two extremes of AI right now. That's kind of what we're seeing. And that's just on the generative AI side. This idea of agentic AI is coming on very strong and is starting to consume all the oxygen in the space. So we're seeing a lot on the agentic side where people are trying to figure this out and figure out how to deal with it, how to orient themselves towards it.
Yeah, I think it's a real concern. I've seen this with people that I work with. It feels like that outsourcing of the mind, it's not good.
No, it's exciting. It's optimistic and there's a lot of optimism but like I said people are terrified because to ignore it is, well, ignore at your own risk, but it's almost heresy in organizations because it's become such a board mandate. Figure out AI. I want an AI strategy. We need AI now. And people don't really know what that means. Is it just good enough to turn on copilot? Am I now AI enabled or do I have to do something bigger than that? On the walk over here, I looked at three presentations that were occurring and they were all AI presentations and they all start the same way. Hey, we're all talking about AI. Everybody's talking about AI and I'm going to give you my AI point of view on whatever the technology happens to be.
But from a cyber security point of view, I mean, Radware [is a] cyber security company, it's a worry.
It's a huge worry. And if I had to boil it down to maybe five things we're seeing with customers, what we're hearing from our customers, it kind of goes like this. We're in the cyber security business. Increasingly, we're in the AI and the API security business, and they're both very much linked technologies.
The first place we start is data. The biggest concern customers have is data. Securing their data, making sure that they maintain control long term over their data. Data is the lifeblood of AI. I talked to my son, my son's a young naval officer on the special operations side, and their big concern about AI is not what it can do for the mission, but accidentally putting confidential information into an AI prompt. And then you've got all sorts of big problems. So it starts with data.
The next thing we hear from customers is identity. The identity increasingly of the agents because sooner rather than later agent equals user. And so everything related to identity becomes a big concern and we haven't really quite figured that out. So it's access control. It's everything that you would do in terms of establishing and validating identity today now extends into that agentic world.
Number three, we see visibility. You can't secure what you can't see. And visibility is a big issue. How many agents are out in my environment? How many APIs are out in my environment? Again, the relationship between AI and API is very closely linked. AIs are enabled and made possible through this very sophisticated world of APIs. So you've got to have visibility into each. And the idea of shadow AI is now the new thing. You've heard shadow IT. So now we're dealing with shadow AI. How many AI tools are out there that I don't even know about?
The fourth thing we hear from customers is the threat landscape. The threat landscape has changed dramatically. You may have heard this. I learned this fact a week or so ago. In the world of phishing, the least attacked country, like almost zero phishing attacks for the longest time, had been Finland. Interesting. Because phishing, the way that I uncover a phishing attack is in mistakes made in language. Like the translation is not great. Yeah, it's a good point. Two years later, that's not the trend anymore. Finland is right up there with everybody else because AI has leveled the game. It's leveled the playing field. It doesn't make mistakes in Finnish born phishing attacks. And so the threat landscape's completely different.
And then finally, number five, consistently what we hear from customers is the loss of autonomy, right? That man in the middle, that human element. These agents operate and reason in an autonomous way. And so we've flipped this whole thing up on its head. And so taken together now it's a whole new approach to cyber security. This is where we find ourselves in AI. And so we at Radware have carved out our own little niche in the market where we bring a lot of expertise and we bring a lot of heritage to innovation. And we happen to see that cyber security opportunity in a little bit different way than some of the other competitors or new players in the industry.
All companies actually deploying agents?
Yeah. No, companies are deploying agents. I will tell you I think, truth in lending, in some cases it's a little bit of a solution looking for a problem. And so companies are trying to figure out what would I do and I'm not so sure that over time the consumer use case doesn't become the prevalent use case versus the enterprise use case. Now the difference between generative AI and agentic AI is outsourcing the ability to think versus outsourcing labor. And so I think over time, this is why I don't believe we're going to replace people in the workforce as it's been predicted at that level. There'll be some reduction and consolidation, but we're going to use agentic AI as a way to help augment the way in which we think, the way in which we do our jobs and that's a great usage of AI.
Are you finding that companies are getting attacked more through AI? Like you use the Finnish example, are you finding that companies are or hackers are using AI to attack companies more often now?
100%. Take the DDoS world. The DDoS world, those attacks are becoming increasingly more sophisticated because of the sophistication in the innovation in AI. So not only are these high volume attacks, but they're innovative attacks. They're very creative. And so now the mission is to stay up. You've got to stay pace with that. So it becomes AI. You're fighting AI with AI. And so yeah, you see that the attacks are new. They're creative. They're innovative and when you think you have one figured out, you've got to orient to the next one because it's going to come at you in a different way. So it is literally AI sort of squaring off with itself in this new cyber security world we find ourselves.
Have you got stories or examples without mentioning names where AI has caused huge problems for companies or where AI has solved a huge problem for a company?
I think there's big stories out there. I don't think we've seen the real benefit of agentic AI as we think it can be. Like we haven't seen it at scale. Let me give you a simple example. Back to that high school education example. I had a conversation with a high school instructor. He teaches criminal justice. 28 years, retired sheriff's department, teaches criminal justice. And I asked him, "Tell me about AI in the classroom." And he said, "Let me tell you, he did the most remarkable thing." He went into AI, ChatGPT, and he created great prompts and he asked the LLM the model to give him a one hour [lesson plan] and he broke it down. 10 minutes of this, 5 minutes of this, 20 minutes of a one hour class on the proper use of handcuffs. It's a criminal justice class. Now he knows how to do this. But he said it generated the most remarkable lesson plan for him in one hour as if it had been doing it forever. And this is a guy who knows his business. And I said, "Was it useful? Did you put this to work?" He said, "I put it to work that day." He said, "I ran the same prompt on how to properly administer a lie detector." And I gave it the same prompt. Said, "I want this much time for this, this much for this." Same thing, put a great lesson plan. So that's a great use of AI.
Sticking to that education example because it's simple. 85% of kids use it today. I gave you that first headline of this being a generational threat and at risk is nothing less than the mind. The concern, and what my daughter who's a high school teacher has said, is that kids want that instant gratification. And when it comes to thinking and reasoning and writing, the learning lies in the struggle. And with AI, there's much less struggle. And so you've heard the old saying, I know what's in my mind, I just can't write it down. And some people may say, you don't know what's in your mind. Because that composition, that writing it down, is very important. So I think that's an example where AI is working against us because kids aren't learning the process of thinking through and reasoning skills and even philosophy. They're going straight to AI.
Increasingly the teachers can see what's an AI generated thing and what's not. I saw a poll or research that was done out of Georgetown that said a human written essay contains eight times more relevant fresh ideas and content than an AI generated paper. Eight times more. And so that's data that we need to understand and consider as we talk about the pros of AI and the cons of AI.
Let me give you a statistic when we talk about people's opinions on AI. It was an NBC News poll that was recently conducted. A thousand people, and they wanted to know people's favorability of AI. Like what do you think? 26% favorable. That's low. 46% unfavorable. But here's the real aha of the poll. AI polled worse than Donald Trump, ICE, and Kamala Harris. That's huge for the US, right? Right. AI polled worse than that and better than the Democratic party and Iran. So this just shows you where people are with AI. You've seen recently the college graduation videos and people getting booed. And this time it was the architects of artificial intelligence. We're using a new AI system as our reader, [and they got] booed in the middle at the mention of AI. So there's a real concern in these college graduates. They've spent four years and a lot of money and studied something and all of a sudden they think that AI is going to take that job, that that's not a job. And so I think somewhere in the middle is where this ends up. I think it's hyperbolic right now. There's just a ton of investment. Mention AI in your earnings report and you'll go up by 25%. Somewhere in the middle is where we end up. But it is becoming an agentic world. Agentic AI is and will be the thing and this is where I think so much innovation will be directed. Now the problem becomes how do you secure it?
Exactly. What does security look like in that world? Because this is a completely different problem. You're dealing with an autonomous agent, something that reasons on its own and takes its own action. And has access to your confidential data. Right, back to that data. Back to the lifeblood. And so the stakes are very, very high.
When it comes to agentic security, the world tends to look at it from the point of view of guardrails. Rules we can create. But what we found is that's insufficient. The agents don't pay attention to the rules. They don't pay attention to the guardrails. It's not efficient. And so the Radware approach has been, we think, radically different. Again, it comes from years of innovation in how we fight very innovative volumetric DDoS attacks, how we fight and resolve bot mitigation, what we do in our web application firewall. So years and years of innovation, the heritage, and we come at this from an intent perspective. So we check and validate intent before execution. So it's behavioral based. We're validating the intent of the agent from a behavioral perspective. So these are the algorithms that speak to our innovation because we don't think guardrails are enough.
And so we come at AI and AI security from two perspectives. Protect AI and serve AI. When it comes to agentic AI protection, it's a behavioral based approach that we take. It's radically different. We think it's incredibly innovative and we think we've got a head start on the market here. As customers and large enterprises start to wake up about this agentic AI security problem, we tell a good story. We have a product today. This is not a simple thing. This becomes complicated work. But again, we are validating the intent, the behavior of the agent to make sure that it's doing what it was designed to do. And that's different than even bot mitigation, bot management, because that's all prompt based. So you're dealing with an autonomous agent and you have to deal with it on its terms from a behavioral perspective.
From the serve AI side, it is making sure, listen, we want to enable the agentic AI economy, not impede it. We don't want to shut it down. We don't want to close it in with guardrails and a bunch of rules. We want this economy to open up and for it to flow within our customers. And so on the serve AI side, it's bringing together everything that we do from DDoS mitigation and bot mitigation and web application firewalls and our entire cyber security portfolio offering that make sure that agentic AI experience occurs and that there's no impediment. So again, protect and serve is kind of how we approach this market. It's still very young. Customers are still trying to figure out, well, the board's telling me I need to use AI. I think it's more than ChatGPT. What's this need to look like? And what would this agentic enterprise look like for me?
So customers aren't deploying it that much yet, right? Not yet. No. But they're looking at it. They're looking at it.
What's hindering it? Sorry to interrupt. I think what's hindering it is there's no formal policy. It's all the things that you would do to create an architecture. And so the underpinnings of that architecture haven't been thought through. There's this big rush to market and everybody's talking AI. And I live in Northern Virginia, not far from Loudoun County and Ashburn where all you see are data centers and it's now flowing into my county in Prince William County. So data centers are everywhere. The infrastructure is being built. I would liken it to the dark fiber phenomena we saw years ago when we were just laying dark fiber just because we had to have it. And so I think this is going to get to a point where it right sizes itself. We've got the Lego blocks, if you will, now it's time to start building this. It's time to start putting this together. But the things that impede it would be the policies. We don't have the policies. I don't know if I even have the right business problems I'm trying to solve. There's a lot of creativity. There's a lot of thinking. In some of the big companies, cloud mythos is a big deal.
Yeah, I was going to ask you about that. So perhaps you can talk about that. Well, look at the alliances that are being formed. Glasswing for example, they're all sort of rallying around and I think they're up to 16 or 20 companies in Glasswing now, including Cisco, that have agreed to go look at their code and fix code which we thought was error free, bug free, and Cloud Mythos has changed that dramatically. And so they're all making this collective commitment now to go improve the quality and the security of that code. So that's a great example of AI doing on one hand a very good thing and recognizing on the other hand the potential for it to be used in a nefarious way. But I think Cloud Mythos has been and will be game changing for the business in a very positive way.
But the problem is if that gets out in the wrong hands or something similar, terrible problem. It's a nightmare. That's a nightmare doomsday scenario. Companies have been, it's the great awakening, like we're having this great awakening. And when I talk about or think about the cyber security imperative for AI, it's a little bit of a whisper right now and I think it goes from whisper to scream very quickly. And at some point here imminently you're going to see the industry screaming about this. We haven't seen one of those big AI attacks yet, like you've seen on the cyber side that we've seen consistently over the last few years, big attacks, whether it's utility or banking or anything in between. And when you see that on the AI side, then I think the industry starts to double down even more. And I think that's how cyber security tends to go. It goes from a whisper to a much bigger scream at some point and I feel like that's imminently ahead of us.
What are attackers using AI for? Is it phishing? Is it writing malware? What are you seeing out there?
It is phishing. It is writing malware. It's writing better phishing attacks in Finnish. And everything in between. It's making these attacks more sophisticated. And they can be launched at scale, at this just unbelievable scale of size and innovation for everything in between. And in the hacker world, I think these types of things are freely shared. So the know how and the expertise and the competency is likewise advancing very, very quickly.
A lot of people watching will know about DDoS attacks. I'm assuming DDoS is still going on like crazy. Listen, I would have thought DDoS, I feel like it's almost a poor man's attack. And but it's becoming so much more than that. It is so sophisticated. It's made so much more sophisticated and innovative through AI and the volumes at which these attacks are being perpetrated. It's a simple thing. Simple, not easy. It's a simple thing to do and it's not particularly easy to protect against. Now, this is what we do. This is our core competency. This is who we are from a hybrid DDoS mitigation perspective. We do it better than anybody. I do it on prem and hardware. I can do it in the cloud. I can escalate to the cloud. So DDoS attack mitigation is here to stay and it's only getting more sophisticated and that's through AI.
When you say sophisticated, are they able to attack it like layer 7? Are they able to create unique patterns that you can't recognize? What are they doing? Everything in between. The layer 7 attacks, the traditional DDoS attacks that we've been dealing with for years, the sophistication within those attacks. So much of it is the undetectability of some of these attacks because the capabilities have advanced dramatically through AI. And I think the way that we go about fighting those attacks again is from a behavioral perspective, [which] sets us apart. It's what's made us successful in the DDoS world. So we've carried that heritage, that know how, that innovation into the AI security world. We approach it the same way and we are equally effective in fighting those attacks. The attacks now and the attacks that we'll see in the future.
What about APIs? Because I'm assuming API attacks are still going on every day. If we want to talk about present time, the biggest threat organizations, enterprises face is API security, period. And the relationship, period, the relationship between API and AI security is very tight. So it's not great to show up and solve one and not the other, which is why we have a solution for both. API security, listen, APIs by and large weren't written with security in mind. They were written quickly and hastily and deployed. They were written to solve a function and they did. The business logic of APIs is what's typically attacked. And so when we talk to organizations, the first question is how many APIs do you have? I don't know, maybe we have a thousand. And when you run a mature discovery tool, you find out they've got 15,000 APIs. 15,000. And then you can start to show the vulnerability that exists in these APIs. And it is breathtaking. It's a breathtaking problem. So they go hand in hand. And nothing about that relationship is going to change between AI and API, agentic AI. And you've got to solve both problems. If I'm an enterprise today and you're not addressing your API security problem, then you have a massive vulnerability you don't know anything about. API security is an enormous problem that most enterprises still don't know they have.
Yeah, I can't say enough about, the API problem is a problem. You can say that loud. It is shocking, the simplicity of the attacks and the elegance is in the simplicity of some of these API attacks that you couldn't believe that the vulnerability at that level exists. And when we go in and we do an inventory and discovery of APIs and then we do a vulnerability assessment, it is eye opening. It should be a board mandate in every company right now. Just API security before you even think about AI and what your agentic AI strategy is going to be. Get your API house in order.
I wanted to ask you, top attacks. Is it DDoS? Is it APIs? Is it AI or all of the above? Perhaps you can give us the top three most attacked things or attack vectors. I think the top three most attacked, DDoS for sure. DDoS is in every one of our customers that we protect, we have great visibility into the attacks that we're fighting off and preventing and it's enormous. API is very close and may well eclipse DDoS because of how much API is out there, how much poorly written vulnerable API is out there. And then from our view of the world it's what a WAF does. It's all the attacks coming, or maybe it's bots. Yeah, it's bots. Bot mitigation. Fighting bots is hard. It's manpower intensive because you've always got to get one step ahead of the attacker and it's data science heavy. It's very heavy data science. So DDoS, API, and bot mitigation from our view of the world are three very, very prevalent attacks that we see not infrequently.
The difference between Radware and a lot of cyber security companies is we are in the fight with our customers every day. I am meaningfully in the fight with my customers fighting DDoS attacks, fighting these API attacks, fighting these massive bot problems and very soon fighting the agentic AI security attack that's right around the corner that we're going to see.
So we've been talking about stuff happening today and in the past, but what do you see coming? If I listen to the jungle drums, I think three things stand out for me. I think this AI bubble is going to burst in a good way. It's not a Y2K size problem, but it's going to right size itself and then we're going to get to work and I think you're going to see some data center consolidation. There's a lot of data center out there. I think we get smarter in terms of processor capacity and so there's more innovation to be had there. Number one. Number two, agentic AI security will become a board mandate for all companies because these agents are going to find their way into our businesses, into our lives, our personal lives, our professional lives, and securing these agents is going to be one of the biggest challenges that we have. And if I had a third prediction in my crystal ball, I think agent becomes synonymous with user. Like an agent is a user. It has an identity. We give it access control. We treat it like a user and we are giving up autonomy. So the loss of autonomy accompanies that. But I think that's the trade off the world's willing to make to simplify, to get the most out of AI and agentic AI. We're willing to give up some of that autonomy for the simplicity and the power of AI. So those three things stand out for me.
You mentioned the students booing, people talking about AI and it seems to be happening quite a lot where kids are not happy. It's our fault. We're developing this stuff and now their future might be limited from what they seem to see. What's your advice to young people? Should they get into cyber security? Should they get into this industry? Is there a future for young people?
I think there's a big future for young people. I think AI will become a meaningfully important part of our lives, but it's new right now and it's scary for all those reasons. I think if you don't come to terms and make your peace with AI, you're going to get left behind. And I think AI becomes something that just augments what we do and how we do it. So on one hand, I don't suggest that we outsource our thinking to AI for all those reasons. I think generative AI will start recycling and generating slop at some point if that's what we do. But an alternative use of AI might be, if you were going to go back in time and you wanted to interview someone from history who's no longer alive, come up with five good interview questions because that's the genius of your thinking. And let AI answer it. And so now have an interview with AI. That's a great way to learn and think. And I think the outsourcing of labor is going to find a meaningful and important part of how we do our work. But I don't think it's going to eliminate jobs at the level and scale that we think it is. But I think these young people have a right to be concerned. The irony is two days ago they were probably using AI to write their term paper. And so this is just, there's a little bit of tough love. And you've seen some of the keynote speakers, graduation speakers, they would respond with a little tough love. Like, hey, I know you don't like to hear this, but here's the reality. Here's the tough love. You're going to be okay. You're going to have to come to terms with it. There's plenty of jobs that aren't going to be replaced or outsourced by AI. And even companies today that say I'm going to do a reduction in force of 10% because of AI, I'm not sure that that's the case. So we're blaming AI on a lot of things and maybe it's an easy, convenient thing to do, but I think AI is here to stay. It's going to right size itself and we will see over time it will make us all better at what we do.
Now if I'm interested in learning more, like say API security or about DDoS, etc. Does Radware have resources that I can read or find? Go to radware.com and you'll have all the resources you want and organized by technology and things that are important to you. So whether it's DDoS or API security or our new agentic AI security, it's all there. There's great technical depth. We do online presentations from time to time. And so that's the best place to go is radware.com.
Randy, thanks so much. I really appreciate it. Yeah, thank you for the time. This has been great. I really appreciate the opportunity.
