Cybersecurity event highlights AI agent risks and the need to limit their authority

At a cybersecurity event involving the Public Cybersecurity Association, the opportunities and risks of AI agents, referred to as "Agentic AI", were discussed. The association's President, Dr. Nurullah Celal Uslu, said the authority granted to these systems must be carefully limited to keep them secure.

The program also included speeches by Prof. Dr. Ergün Eraslan, Dean of the Faculty of Engineering and Natural Sciences at Ankara Yıldırım Beyazıt University; Alpaslan Kesici, General Coordinator of the Turkey Cybersecurity Cluster; and Mustafa Gençer, Netsys Ankara Regional Manager.

Agents plan and make decisions

Uslu noted that AI technologies are advancing rapidly. Unlike classic generative AI systems, he said, agents can plan toward the goals they are given, use various tools and take independent decisions. He therefore said a wrong decision could have direct consequences on real systems.

Yapay zekaya ne kadar yetenek kazandırdığımız kadar, ona ne kadar yetki verdiğimiz de önemlidir.

Nurullah Celal Uslu

Uslu stressed that limiting authority, obtaining human approval for critical actions and carrying out independent security checks are important for securing AI systems.

Hands-on training and competition

Uslu said the training was not only theoretical: participants saw how AI agents can be used in cybersecurity, built a security agent, and examined how prompt injection attacks can affect an agent's behavior and which precautions can be taken against them. In his view, merely using or developing AI will not be enough in the future; developing secure AI systems will be an important field of expertise.

The event ended with a competition with prizes for participants. Winners received academic publications and one year of access to a cybersecurity training platform provided by Hackviser.