AI and cybersecurity dominated the conversation at ESET World 2025. ChannelPro sat down with ESET executives during the conference: Juraj Jánošík, director of automated systems and intelligent solutions, and Filip Mazán, senior manager of advanced threat detection and AI research.
They discussed artificial intelligence, cybersecurity, and how MSPs can capitalize on both. Here are the highlights from what they shared, edited for clarity and brevity.
How Does AI Help MSPs with Things Like Endpoint Protection?

Juraj Jánošík
Jánošík: Endpoints are a great place to start. Our endpoint protection capabilities are very strong. We develop really optimized detection models running directly on the endpoint. That means the amount of data sent to a centralized console is much lower.
With our MDR service, which is built on top of our EDR, we have this AI-powered incident creator that helps take a lot of the workload off the MDR security analyst. We put a lot of effort into training these models to reduce false positives and to really focus on showing users and analysts only the incidents that matter most.
Mazán: These AI models can really bring the number down. In some cases, up to 50% of incidents don’t even need to be looked at by our analysts because they’re identified as noise.
What Should MSPs Understand About the Convergence of AI and Cybersecurity?
Mazán: It’s not a symmetrical fight. Attackers are using generative AI nowadays to create scam campaigns, phishing attempts, and maybe even deepfakes. We use different, more specialized AI models that are good at their single responsibility like detecting malware. It isn’t the same technology.
Jánošík: The models we use are trained by us on our data. We have team of people figuring out and refining our models. When people think of AI, they think of ChatGPT … but what we have here are purpose-intended technologies.
If MSPs Are Shopping for AI-powered Cybersecurity Tools, How Do They Know What’s Real vs. Marketing Hype?
Jánošík: Consider independent testing laboratories, like AV-Comparatives, AV-TEST, and MITRE, among others. The testing results will give you insight into what the companies are claiming in their marketing materials.
Mazán: I also suggest checking out published papers because that analysis can show a lot.
Jánošík: Almost every year, we publish white papers about how our technology works. We try to be transparent when we say that we are using this model or that technology.
Is Your Team Allowed to Use ChatGPT at Work?
Jánošík: Yes, they are. We also have our own private implementation of ChatGPT. But our people are trained on how to use it responsibly and what to not put there, such as client data.
Mazán: Also no source codes are uploaded. We use LLMs for counseling and mentoring approaches. We don’t use it with sensitive data.
Jánošík: If you ask ChatGPT something, check the results. There needs to be a human in the loop. The model will give you a really persuasive answer, but it’s not always correct.
What’s the Biggest Challenge in Integrating AI into Cybersecurity Services?
Jánošík: You need people with AI engineering and research knowledge — people who can build the data sets, run the hardware, and keep everything maintained. When it comes to generative AI, we were looking for partners running in the cloud but with a private space. Nothing gets logged. Privacy is what matters; trust is the key.
Do You Think MSPs Risk Becoming Too Dependent on AI Tools?
Mazán: Yes, it’s a slippery slope because people start to automatically accept everything. We need to be really cautious. There’s also an accountability problem. If the model makes a bad decision, who is ultimately responsible?
Jánošík: Imagine you’re providing security for a hospital client and suddenly, there’s a false positive. The MSP reacts and cuts the network — just in case. That could end really badly.
Mazán: Different types of problems come up when you start using it. Instead of writing your code and thinking it through logically, you just start accepting the GPT output. You become a code reviewer and end up reviewing the mistakes the AI made. You lose your ability to be creative. I don’t think is a great approach.
How Should MSPs Train Their Staff on AI Tools?
Jánošík: We have a new training program focused on operators and analysts. We’re working to build out these awareness and educational efforts. At least half of the challenge in solving security issues isn’t technical. It’s education and awareness.

Filip Mazán
Will We One Day Wake Up to News that AGI (Artificial General Intelligence) Has Arrived?
Mazán: The companies leading in AI are still trying to scale up but that can only go so far. At some point, there has to be some kind of breakthrough. There’s work in progress, but right now, it doesn’t seem possible.
Jánošík: I’d even say we might be on the road to a dead end. Running these models is extremely energy-inefficient. You need specialized hardware which is very expensive. Also, GPT models don’t mimic how the human brain works. When a child learns what a cat or a dog is, they don’t need to see thousands of images to understand it. But that’s how GPT models work; it’s a statistics-based model. You need to show thousands of pictures so they can distinguish the cat from the dog.
Should MSPs Talk About Their AI Capabilities when Selling Their Cybersecurity Services?
Jánošík: It’s definitely it’s important to mention it because it’s such an important topic right now. Talk about how effective it is, where the limits are, what are the potential security risks are. That way the user can decide how heavily they rely on it.
Every time we use generative models in our products, we make sure the response is as consistent and correct as possible within the limits of the technology. We mark everything in the user interface that’s AI generated with a disclaimer: “This was generated by AI, so you need to evaluate it.”
Remember, everyone thinks of generative AI as driving higher effectiveness but studies show the increase in productivity is only about 5%. We came to 6% in our research, so it’s pretty close. Unlike specialized AI tools, the ROI of generative AI is really hard to define. Your customers should remember that.
Learn More
➡️ Dive into the ESET World’s content from Day 1 and Day 2.
⏩ Read about the most important MSP-focused announcements from ESET World 2025.
⏭️ Check out our new ESET World 2025 article: How to Make Cybersecurity Practical
Images: iStock, ESET