Jakarta, INTI - The rapid development of digital technology brings new challenges to the field of cybersecurity. Although various artificial intelligence (AI) methods have achieved very high levels of accuracy in detecting cyberattacks, researchers are reminded to focus not only on numerical results, but also on the system's ability to deal with evolving threats in the real world.
This was conveyed by Prof. Grzegorz Kolaczek, Head of the Department of Computer Science and Systems Engineering in the Faculty of Information and Communication Technology at Wrocław University of Science and Technology.
On Tuesday, July 14, Kolaczek visited the B.J. Habibie Building in Jakarta as a guest lecturer with researchers from the Research Center for Artificial Intelligence and Cybersecurity (PRKAKS) of BRIN.
Modern Cybersecurity Research Needs New Approach
In his lecture, Kolaczek addressed the challenges of modern cybersecurity. He explained that many studies have competed to produce intrusion detection models with accuracy rates approaching 100 percent. However, he argued that these achievements do not necessarily reflect the system's ability to handle real-world conditions.
"High accuracy may seem impressive, but the question is whether the model can actually perform on real-world systems that constantly change. That's the real challenge in cybersecurity," Kolaczek explained.
He explained that most cybersecurity research still uses decades-old public datasets as the primary reference. While useful for comparing different approaches, these datasets are no longer considered to fully represent the evolving characteristics of modern computer systems.
Kolaczek assessed that developments in AI technology, including convolutional neural networks (CNN), Long Short-Term Memory (LSTM), transformers, autoencoders, and graph neural networks, have made significant contributions to the advancement of cyber threat detection systems. However, he emphasized that the primary goal of research is not simply to create new algorithms.
"As scientists, our goal is not simply to obtain the best results on a specific dataset, but to produce solutions that can actually be implemented to address cyberattacks in operational environments," he emphasized.
Data Quality is Important
He further explained that data quality is a crucial factor in determining the success of AI-based cybersecurity systems. The current challenge is no longer a lack of data, but rather ensuring that the data used is truly accurate, relevant, and reflects current conditions.
In addition to data quality, changes in user behavior, developments in network technology, and the evolution of attack patterns also require cybersecurity models to be able to adapt continuously. Systems that rely solely on static models risk losing effectiveness when faced with new threats.
Kolaczek also highlighted the unique characteristics of cyberattacks. Unlike fixed threats, attackers are constantly learning about defense systems, adapting their strategies, and finding new vulnerabilities to evade detection. Therefore, cybersecurity research needs to consider the dynamics of attacker behavior as part of the model development process.
As a direction for future development, Kolaczek introduced several approaches considered promising, including adversarial training, certified robustness, ensemble learning, and federated learning. These approaches aim to increase model resilience against various forms of data manipulation and increasingly complex attacks. He added that although these methods have been widely developed, further research is needed to effectively apply them to the dynamic modern cybersecurity environment.
Conclusion
Prof. Grzegorz Kolaczek from the Wrocław University of Science and Technology emphasized that artificial intelligence (AI) research for cybersecurity should not only focus on high accuracy but also ensure models are capable of addressing real-world challenges. He noted that many research still relies on outdated datasets that are poorly representative of modern systems, making data quality and relevance crucial in developing cyberattack detection systems. Kolaczek emphasized that AI models must be able to adapt to changing attack patterns, user behavior, and network technology.
Read more: Poltek SSN Developed ASICK AI Chatbot to Support Bogor Regency Diskominfo’s Public Information Services