VIDEO — 20 Nov 2025

AI and cyber threats, security tools you can build yourself and startup ideas

A talk at DSC Europe 25 in Belgrade: AI in cyber attacks, security tools anyone can build, startup ideas and digital violence among children.

Published in

DSC Europe 25

Role

Speaker

Watch on YouTube

At DSC Europe 25 in Belgrade’s Sava Centar, I gave the closing talk of the fourth day in the track on AI in cybersecurity. The announced topic was how AI is redefining cyber threats, and I approached it from an entrepreneur’s perspective: how generative AI is used against companies and ordinary people, which security tools you can build yourself and which problems are still waiting for a startup. I also talked about AI beyond security and digital violence among children, and took questions from the audience.

AI and security from an entrepreneur’s perspective

It was the last session of a long day, so I kept the talk light. I focused on small AI tools anyone can build to improve security at work, and on startup ideas, because cybersecurity still has a lot of unsolved problems, gaps and missing tools.

Where I come from and what I do

I have been in cybersecurity for about 13 years, in consultancy and in many companies in Europe, and I have lived in Germany, Abu Dhabi and New York. I am originally from Montenegro, I currently live in Belgrade and run a few businesses, and I help young entrepreneurs in Serbia and Montenegro launch new ideas.

UN1QUELY, founded in Novi Sad, has provided cybersecurity services for five years and has an elite team of penetration testers. Secfix and Skenify are SaaS product startups that we also started. The UN1QUELY Academy is completely free for people entering cybersecurity, with recorded training in Serbian in security management, offensive security and penetration testing, and defensive security operations.

AI is a tool, and it matters who is behind the wheel

AI is a tool, and in cybersecurity the question is who is behind the wheel and with what intention. The cybersecurity industry exists at all because some people have bad intentions. Anything you can generate, you can also generate for a malicious purpose, usually social engineering: phishing emails tailored to one person, deepfake videos that get more realistic with every new model, and voice mimicking, which has already been perfected.

These attacks now target the general public more than large enterprises. I have received many smishing messages from a fake postal service asking for my card details. On the morning of the talk, an email reached my HR with the full name of one of my employees, asking in Serbian Cyrillic for his salary to go to a new bank account. AI makes such attacks easier and faster.

What you can build with AI today

Your own security tools in a few prompts

I had already tested the newly released Gemini 3 on small tools for my own logs. Export logs from Microsoft Intune or Microsoft Defender, ask such a tool to correlate them and point out anomalies, errors and critical issues, and it works like a SIEM. It takes two minutes to build, and with Lovable you can make it a whole web application connected to your tools through APIs.

The security industry presents fancy new AI tools, although they are the same tools with AI agents built in. I shouldn’t really say this, since I sell security tools myself, but you probably don’t need all of them, and a very small company can build its own set with enough vibe coding.

Detecting deepfakes and AI slop

Any tool that can detect a deepfake with 100% certainty will be a new unicorn. A year ago, hands in AI videos still had six or seven fingers, but when I recently generated ordinary scenes from real photos with Veo 3.1 and Sora 2, about half the time I could not tell anything was wrong. Tools that recognise AI content and stop algorithms from pushing AI slop higher in our feeds would be a huge opportunity.

Automated penetration testing

Penetration testing is a great process, but it is manual, expensive and captures a single point in time. I know, because we sell it: two people need at least 5–10 days to test a system, and every new feature means testing again.

My team should rely on automated, continuous testing that keeps trying to break in and reports what it finds, from a back door to administrative privileges gained with a basic user account. Penetration testers would then verify the findings in about 20% of the time they spend today, and whoever builds that platform will make big money.

Compliance and quantum-safe encryption

Serbia has just adopted a new information security law with a lot of new requirements, and most companies don’t know where to start. I can bet that if you sit down with Lovable tonight, download the law and build a project-management tool for implementing it, by 2 a.m. you will have an almost ready product for the Serbian market, and tomorrow you can sell it as a subscription. It can be an easy one-person startup, because compliance is mostly project management. I won’t compete with you, since we already have a different tool on the German market.

Quantum-safe encryption has been a topic for two or three years, in preparation for the day when attackers can use quantum computers to break current encryption. The financial sector will be the first target, because attackers always go for the money, so regulatory requirements in the European Union and other countries are already forcing it to move to quantum-safe methods.

Beyond cybersecurity: healthcare, education and the environment

Startups and large companies are investing heavily in using AI to solve problems at scale. Follow that news and watch less of the depressing stuff, because in cybersecurity we always talk about threats, risks and attacks.

—Healthcare. Predictions say AI will solve every single problem in healthcare and, together with quantum computing, let us produce any protein or drug tailored to each person. What would take 100 years without AI, we will probably do in a decade.
—Education. My most used tool is Google’s NotebookLM, which quizzes you on the material you upload or turns it into a podcast for the car. My biggest recommendation is Khan Academy, a completely free online school whose AI tutor speaks every language and keeps quizzing children until they master the subject, without giving ready answers.
—The environment. Just as a system can track your heart rate and temperature, AI with IoT devices around the world could feel the pulse of entire ecosystems and show where the soil, the air or the biosphere is at risk before it is too late.

Digital violence among children

Teenagers in schools are bullied online, recorded on TikTok and targeted with AI deepfakes made to mock them, and they feel shame even when the video is not real. They become depressed and left out, and in the worst case this ends in suicide, which is already happening around the world. Many of us can use these same tools to help children be safer online.

That is why I see cybersecurity as a very good calling: you can protect companies from cyber attacks and also help keep children safe online from predators and bullies.

Stay curious, creative and ethical

How you apply AI is up to you, and my one suggestion is not to become a cybercriminal. With AI, you should stay curious.

Creativity is the number one thing that has to stay human, and it always will. When you ask ChatGPT for ten ideas, two or three will be great and the rest rubbish, and you are the one who recognises which are creative, because ChatGPT only gives statistically relevant suggestions. Creativity comes first in every field I talked about, and you should always stay ethical.

Questions from the audience

Asked about the harm AI does to the environment, I agreed that the problem is real and pointed to work already under way on moving data centres to low Earth orbit, where they would need no water or active cooling and could run on solar power. It sounds like science fiction for now, but I am always optimistic.

Asked about lateral movement during an attack, I said that in a company without any security solutions, an attacker who owns an administrative account has pretty much won. AI can correlate logs faster than any analyst and stop the attack before a human notices, and SIEM tools plus next-generation malware protection are the bare minimum.

More conversations & talks