I'm preparing to deploy Defensepro devices to our network, but I have a few questions and need expert advice and accurate information about Radware's DDoS protection solution.

Our network model is shown in the attached image, including two ISPs, and we use BGP to redirect traffic. We plan to place two Defensepro devices on two separate lines. However, because we use iBGP, there might be cases where connections enter one line and exit through the other. Therefore, please help me determine the optimal placement of the two DDoS protection devices and whether asymmetrical traffic will be dropped by the DDoS devices.

Short version:

Teams had API keys in their code. Mostly for simple things like billing or basic services .Then they enabled Gemini API.

Using the same keys.

But now they could:

• Access new endpoints

• Interact with AI services

• Reach data they were not meant to

• Generate unexpected costs

No exploit.
No code broken. Just a change in behavior.

What’s interesting here is not the AI itself.

It’s the assumption: “this key is not sensitive”

That assumption changed.
But the controls didn’t.

This is very close to what we see in application logic issues:

• permissions that grow over time

• APIs that behave differently after changes

• things that are “technically allowed” but shouldn’t happen

We put together a quick poll to understand how people handle this in real environments:
👉 [Link to poll]

We’ll share the results here once we collect enough responses.

Curious to hear how you approach this today.

Stay safe and proactive

Whenever we talk about major breaches, we usually imagine hackers finding a bug in the code or a hole in the firewall. But this time, I want to invite you to dive into something completely different—an angle we haven't really explored at Radware until now.

Think about this: $1.5 billion stolen in 156 seconds. Without breaking a single line of code.

How? The Lazarus Group didn’t look for a technical flaw. They simply understood how the business worked—and used that business logic against the organization. It’s scary because it means the vulnerability isn’t in our firewall, but in the way we built our processes.

We’ve invited investigative journalist Geoff White and our own Uri Dorot for a 30-minute conversation. They are going to "deconstruct" the Bybit heist and explain exactly why the defenses most of us build simply wouldn’t have stopped it.

This isn’t a sales pitch. It’s a journey into the adversary’s mind to understand where we are truly exposed.

I’ll be there, and I’d love for you to join us: Click to register.

Sharon.

Have you caught the news? $285 million stolen in just 12 minutes. But the real horror isn’t the speed, it’s the fact that the attack didn’t start with a technical hack. It started six months earlier, at a crypto conference, with a handshake.

While we’re all busy looking for sophisticated bots or AI-generated code vulnerabilities, North Korean hackers (UNC4736) went back to basics: building relationships.

They realized it’s much easier to "hack" a human than it is to hack a system. This was incredibly sophisticated "Low-Tech": they sent actual people to industry events, deposited $1 million of their own money to build credibility, and spent months in Telegram groups having substantive professional conversations. They didn't look for a "hole" in the code; they simply manipulated authorized users into changing security rules and removing critical protections like the time-lock for them.

The Insight

This story reminds us that security has many faces, and the threat has shifted from "systems" to "minds." Powerful technology is critical, but it must work hand-in-hand with human alertness and systems designed to prevent "human factor" errors.

The Question

Social engineering like this exploits our most valuable asset - trust. If you were in the shoes of the security team at Drift, how do you think you could have defended against or prevented an attack that begins with a simple handshake?

I want to invite you to an exclusive Masterclass designed for those of us facing the daily challenges of digital protection in an ever-evolving world. As part of our Radware University initiative, we’re hosting a deep dive into AppSec—a critical element in keeping our digital ecosystems secure.

The Details: We’re meeting on Thursday, June 11, 2026, at the Cardo Hotel in Rome (Viale del Pattinaggio, 100).

Why join us? The massive shift to the cloud and expanding attack surfaces mean we need a new level of focus on AppSec. This session isn't just about theory; it’s about sharing strategies and best practices to defend applications and APIs from advanced threats.

What we’ll cover:

• AppSec Threat Landscape: A look at emerging threats and how AI is fundamentally changing application security.

• 360° Protection Model: A deep dive into WAAP, Bot Management, Client-Side Protection, and Web DDoS Protection.

• API Security: Since APIs are now integral to exposed web apps, we’ll focus on understanding their vulnerabilities and how to protect them.

• Interactive Workshop: Hands-on technical exercises and real-world scenarios to build operational skills.

Who should attend? This is for the pros—Cybersecurity professionals, IT managers, network architects, security analysts, and sales engineers who want to stay ahead of the curve.

I’d love to see you there. To secure your spot, please register by May 29, 2026.

Click here to sign up: [Link]

What the Community Says

OWASP — the leading global authority on application security — recently published the OWASP Top 10 for Agentic Applications (2026), outlining the ten most critical risks associated with autonomous AI systems. This list is built by a worldwide network of security experts, researchers, and industry contributors, and is quickly becoming a cornerstone reference in the field. If you’d like to dive deeper into the community-driven source itself:
https://genai.owasp.org/resource/owasp-top-10-for-agentic-applications-for-2026/

What Radware Adds

While OWASP maps out the risks, at Radware we focused on the practical side:
How do you actually survive these risks?

In our new guide, How to Survive the OWASP Top Ten for Agentic Applications, we break down each key risk and translate it into real‑world, actionable defense strategies:

✅ How attackers already exploit agentic systems
✅ What real misuse looks like in production environments
✅ How to detect, mitigate, and block these threats in real time

Our goal is to help teams understand not just what the risks are — but what to do next.

What’s Next?

If you're working with AI agents, experimenting with agentic workflows, or building autonomous capabilities:

🔹 Read what the community is saying to get the full landscape
🔹 Read our guide to learn how to defend against these threats effectively

And don’t hesitate to ask questions.
This domain is evolving daily, and we all benefit from sharing knowledge, challenges, and hands‑on experience.

Let’s build safer, smarter agentic AI — together. 🌟

Stay safe and proactive.

The transition from "chatting" with AI to letting AI "act" is a massive leap in productivity—but it also changes the rules of security.

I recently published a blog that captures this shift. Traditional LLMs are like advisors; they give you information, and you decide what to do with it. Agentic AI, however, is like an intern with a set of keys. It can browse the web, access APIs, and execute tasks autonomously.

The risk isn't just about the AI being "smarter"—it's about the Autonomy Risk. When an agent can make decisions without a human in the loop for every step, a single prompt injection or a flawed instruction can lead to real-world actions that are hard to undo.

Read my blog.

Dror