For the past three years, society has been preoccupied with one central question: What impact will the rapid rise of artificial intelligence—especially so-called large language models (LLMs)—have on our lives? What will it mean for children who learn in school with or through LLMs? And how might these technologies reshape the everyday experiences of older generations?

These questions also resonate in the field of digital security. Will LLMs become a true game changer—or will the hype subside and “business as usual” return? One thing is certain: cybersecurity is undergoing significant change, though not always in the way many expected.

LLMs have quickly proven themselves as tools for generating highly convincing phishing emails, fake executive calls, manipulated videos and sophisticated disinformation. These risks are substantial but, in principle, manageable. The real test of LLMs as potential cyberweapons, however, is happening in the more technical corners of the digital realm.

At the heart of cybersecurity lies the task of identifying and eliminating vulnerabilities—programming errors that can allow attackers to execute commands, encrypt data or steal information. Detecting such weaknesses requires deep technical expertise. Here, LLMs offer a powerful productivity boost thanks to their ability to answer complex questions and propose useful solutions—hallucinations notwithstanding. Who benefits? Anyone who uses them, regardless of their intentions.

A similar development can be seen in a very practical domain: hacking. This involves overcoming protective barriers designed to safeguard the integrity of IT systems. To simulate attackers’ behaviour and address security issues quickly, organisations rely on penetration tests. Here too, LLMs increase efficiency and reduce costs—not only for defenders, but also for attackers.

Conclusion: Even in the era of LLMs, the race for technological advantage remains unchanged. Security continues to be a competition between human unscrupulousness—and human integrity.