Skip to main content
4–6 March 2026Adversarial AI

From Poison to Antidote: AI attack and defence training presented at ICISSP 2026

Christoforos Ntantogian presented the lab's work on teaching adversarial AI through hands-on attack and defence exercises at the 12th International Conference on Information Systems Security and Privacy in Marbella.

Conference
ICISSP 2026 (12th edition)
Location
Marbella, Spain
Dates
4–6 March 2026
Presenter
Christoforos Ntantogian
Photo: NITRO project / Ionian University

Christoforos Ntantogian (Dadoyan) of Ionian University presented the paper 'From Poison to Antidote: Advancing Cybersecurity Education with AI Attack and Defense Training' at the 12th International Conference on Information Systems Security and Privacy, held in Marbella, Spain from 4 to 6 March 2026.

The title captures the argument. Machine learning models introduce a class of vulnerability — data poisoning, evasion, model extraction, adversarial examples — that most security curricula still treat as theory. The paper's position is that these attacks should be taught the way buffer overflows are taught: by having students carry them out in a controlled environment and then defend against them.

Part of a wider programme

Adversarial AI is one of the defining features of the NITRO cyber range, which is among the first ranges to emulate attacks on AI models for training and security testing, using open datasets including DeepSlice and Secure5G for 5G scenarios and WSN-DS and CICIDS18 for IoT networks. The aim is to raise awareness of how AI models can be attacked and how robust AI can defend against those attacks.

It also connects to ANTIDOTE — AI Attack and Defense for the Smart Healthcare — a Marie Skłodowska-Curie network launched on 30 January 2025 in which the lab participates, addressing emerging security threats arising from the integration of AI in healthcare.