Pietro Liguori
Researcher in software reliability, cybersecurity, and AI code generation.
University of Naples Federico II
Member of DESSERT Lab
My research sits at the intersection of artificial intelligence and software security. I study how large language models can be used — and how they can be attacked — to generate, analyze, and secure software systems. Before moving to AI-based code generation, I worked extensively on fault injection and failure analysis in large-scale cloud infrastructures.
I earned my Ph.D. at the University of Naples Federico II under the supervision of Prof. Domenico Cotroneo, and I spent nearly a year as a visiting research scholar at the University of North Carolina at Charlotte, working with Prof. Bojan Cukic on AI-based code generation for offensive security.
What I work on
AI Code Generation
Evaluating correctness, robustness, and security of code produced by LLMs and neural code generators. Benchmarks, metrics, and testing methodologies.
Offensive Security via AI
Generating exploits, shellcode, and PowerShell attacks from natural language. Understanding what AI can — and should not — produce.
Software Reliability
Fault injection, failure mode analysis, and runtime detection in large-scale cloud platforms such as OpenStack.
Vulnerability Detection
Static analysis and prompt-based methods for detecting and patching vulnerabilities in both human-written and AI-generated code.
Code, datasets, and models are released on GitHub (DESSERT Lab) and Hugging Face (OSS-forge).
A brief origin story
Legend has it I am the result of a long-running laboratory experiment conducted by Prof. Domenico Cotroneo and Prof. Roberto Natella, somewhere in the basement of the DESSERT Lab.
For years they applied every fault injection technique they had at their disposal — deadlines, rejections, conference reviews at 3 AM, paper #4 of the month, PhD defenses, missing GPUs, broken pipelines, and the occasional Reviewer 2. Against all odds, the subject exhibited remarkable resilience, and the prototype was eventually deployed in production at DIETI - University of Naples Federico II.
Residual failure mode: a deep, possibly unhealthy, passion for finding ways to break systems — especially AI.
Scientific output
Selected recent publications
-
Reading between the Lines: Context-Aware AI-based Generation of Software ExploitsEmpirical Software Engineering, Springer, 2026
-
Will It Break in Production? Metric-Driven Prediction of Residual Defects in Python Systems56th Annual IEEE/IFIP International Conference on Dependable Systems and Networks (DSN 2026)
-
Human-written vs. AI-generated Code: A Large-Scale Study of Defects, Vulnerabilities, and ComplexityIEEE 36th International Symposium on Software Reliability Engineering (ISSRE 2025)
Recent highlights
Selected for a tenure-track position at DIETI, and national qualification as Associate Professor
Selected for a tenure-track researcher position at the Department of Electrical Engineering and Information Technology, University of Naples Federico II, alongside the Italian National Scientific Qualification (ASN) for Associate Professor in both Computer Science and Computer Engineering.
Program Co-Chair — FORGE 2027 Data and Benchmarking Track
Co-chairing, with Zhou Yang (University of Alberta), the Data and Benchmarking Track of the 4th ACM International Conference on AI Foundation Models and Software Engineering, co-located with ICSE 2027 in Dublin. Submissions due 15 November 2026.
Invited lecture at LLMA4SE 2026 — Cáceres
Taught Workshop 3, "Benchmarking the quality of AI-generated code," at the International Summer School on LLM-based Agents for Software Engineering.
Program Chair — DSML 2026
Chaired the 9th Workshop on Dependable and Secure Machine Learning, co-located with DSN 2026.
1-Star MBDA Innovation Award
Recognition for the Artificial Firmware Designer Assistant project — applying generative models to firmware design in high-criticality industrial contexts.
Launched Redlyne — security for AI-generated code
Released Redlyne, an open-source VS Code extension that detects vulnerabilities in AI-generated Python code and proposes one-click patches — running entirely on the user's machine. VS Code Marketplace · GitHub.
Invited talk at USI Lugano
"Residual Faults in the Age of AI-Generated Code," hosted by Prof. Gabriele Bavota at the Università della Svizzera Italiana.
Selected to present DeVAIC at innovIT — San Francisco
Selected by the Italian Innovation and Culture Hub (innovIT) to present DeVAIC — a tool for security assessment of AI-generated code — in San Francisco.
Invited member, IFIP WG 10.4
Invited to the IFIP Working Group 10.4 on Dependable Computing and Fault Tolerance.