AI-powered penetration testing with Shennina: from simulation to validation
An open-access ARES 2024 paper examining AI-driven red teaming — taking automated offensive tooling from simulated environments through to practical validation.
Image: Shennina project (GitHub: mazen160/shennina, README banner)'AI-Powered Penetration Testing using Shennina: From Simulation to Validation', by Stylianos Karagiannis, Leonidas Agathos, Emmanouil Magkos and Christoforos Ntantogian with colleagues, was published open access in the ARES 2024 proceedings.
The paper explores AI-driven red teaming, following the tooling from simulated environments through to practical validation. That progression is the whole argument: automated offensive tools perform impressively in the controlled conditions they were tuned on, and the interesting question is what survives contact with a system nobody designed for the demonstration.
Automation and its limits
Automated penetration testing promises to widen access to a scarce skill. It also risks producing confident reports about attack paths that do not exist. Validating simulation results against real targets is how that risk is measured rather than assumed — and it advances automated cybersecurity testing on an honest footing.
The work belongs to the lab's wider adversarial AI programme, which runs through the NITRO cyber range's AI attack exercises and the ICISSP 2026 paper on teaching AI attack and defence.