Agentic offensive security
Offensive agents that plan and execute real attack chains, ranking #1 among AI teams in security CTFs, competing in Attack/Defense battlegrounds and testing real systems, from OT networks to consumer robots.
Francesco Balassone
AI Security Research Lead @ CYBER RANGES · PhD @ Federico II
I'm Francesco, an AI cybersecurity researcher. I build and evaluate offensive and defensive AI agents that hack, defend and emulate real adversaries, for military, government and enterprise.
An Offensive Agent moves through a network from left to right; a Defensive Agent detects and contains it.
Where offensive security meets AI agents.
Home lab
I'm a PhD student at ARCLAB, in the Department of Electrical Engineering and Information Technology (DIETI) of the University of Naples Federico II.
Inside ARCLAB we run DARKLAB, the lab's cybersecurity branch, working on offensive and defensive AI, adversary emulation and CTFs, a few steps from the Maradona stadium in Fuorigrotta.
Offensive agents that plan and execute real attack chains, ranking #1 among AI teams in security CTFs, competing in Attack/Defense battlegrounds and testing real systems, from OT networks to consumer robots.
AI agents configured as real APT groups such as APT28, APT29, APT41, APT44 and Lazarus raise a hard question: if anyone can emulate a playbook, do TTPs still tell us who attacked?
Static benchmarks saturate. I work on harder yardsticks: meta-benchmarks, harness comparisons and Dynamic Cyber Ranges where LLM defenders fight back in real time.
Are AI agents better at attacking or defending? Across 23 Attack/Defense CTF battlegrounds, defensive agents patched far more often than offensive agents gained initial access (54.3% vs 28.3%), but the advantage vanished once defenders had to keep services up and stop every intrusion.
arXiv:2606.07158
arXiv:2604.24184
arXiv:2603.08665
Proceedings of the AAAI Symposium Series 9(1)
arXiv:2601.05887
AAAI-26 Workshop on AI for Cyber Security (AICS)
AAAI-26 Workshop on Reproducible AI (RAI)
AAAI-26 Workshop on AI for Cyber Security (AICS)
ITASEC & SERICS 2026 · Joint National Conference on Cybersecurity · Cagliari, Feb 9 · recording ↗ (in Italian)
Co-founder & Captain
The competitive hacking team of the University of Naples Federico II.
napwnli was born in 2024 from the finalists of the Naples team of CyberChallenge.IT. I co-founded it and I captain it: we play CTFs all over the world, learn together, and build a community of people who enjoy hard problems like web, pwn, crypto, reversing and Attack/Defense.
PROUD TO BENAPWNLI
First posts are on the way.
I lead AI security research at CYBER RANGES and I'm an industrial PhD student at the University of Naples Federico II.
At CYBER RANGES I lead AI security research, drills and training programs for military, government and law enforcement. My PhD, “Towards Autonomous Red Teaming: Agentic AI for Offensive Cybersecurity Operations”, is supervised by Profs. Simon Pietro Romano and Gaetano Perrone, in partnership with Alias Robotics and CYBER RANGES.
Before that, at Alias Robotics, I co-created CAI and led research on agentic AI for offense and defense: from Jeopardy CTFs to full Attack/Defense competitions, synthetic APT campaigns and production-like cyber ranges.
Before the papers there were CTFs. I co-founded and captain napwnli, and I've gone from CyberChallenge.IT student to instructor, still the best way I know to learn how systems break.
“TL;DR I like computers. I'm not sure they like me…”
Let's talk, research, collaborations, CTFs.