Description

08:00 - 09:00 WELCOMING and Badge collection

09:00 - 09:25 Opening KEYNOTE - Anonymous Speaker (EN)

09:35 - 10:15

• Track 1: Le produit entre la qualité et l'Over-Engineering, Jihène Mejri (FR)
• Track 2: Au secours ! Mon manager me demande des KPIs !, Geoffrey Graveaud (FR)

10:25 - 11:55

• Track 1: (40 mn) Optimizing LLM Inference on GKE for the Rest of Us, Abdel SGHIOUAR
• Track 1: (40 mn) Data Connect : Le nouveau service Firebase de base de données relationnelle, entièrement géré par Google Cloud SQL et reposant sur PostgreSQL, Renaud Tarnec (FR)
• Track 2: 10:25 - 11:55 Accessibility Testing with Chrome DevTools, Josefine Schaefer (EN)

12:05 - 12:40

• Track 1: LLMs White Hat Hacker, Malik Bouhou (FR)
• Track 2: Stratégies de rendu et Angular : passez vos applications au niveau supérieur, Jimmy Kasprzak (FR)

12:40 - 13:30 LUNCH 🥪🥤

13:30 - 14:10

• Track 1: État des Lieux de la Souveraineté du Cloud : Des Clouds Publics aux Clouds Confidentiels, Seifeddin Mansri (FR)
• Track 2: Android + MediaPipe + Gemma 3: Building offline AI-powered apps, Tom Colvin (EN)

14:25 - 15:05

• Track 1: Gemini 2.5 & Gemma 3: What's New, What's Awesome, and Which one to Choose?, Ilias Papachristos (EN)
• Track 2: Fugu: Possibilities with Web Capabilities, Rowdy Rabouw (EN)

15:15 - 16:00

• Track 1: Libérer la puissance des systèmes multi-agents avec Flowise, Laurent Grangeau (FR)
• Track 2: Building a Modern Data Strategy on Google Cloud: From Foundation to AI-Powered Insights, Alejandro De La Cruz López (EN)

16:10 - 16:35 WRAP UP KEYNOTE - Anonymous Speaker (EN)

More information

Malik Bouhou, Research engineer in Cybersecurity, CETIC, will present a study in two phases exploring the potential use of Large Language Models (LLMs) in supporting the development of penetration test scripts.

We will explain the exploration phase where we experimented with multiple LLMs such as chatGPT variants but also local models using Ollama such as llama3, qwen2, gemma2, etc... These LLMs were experimented with different use cases that were chosen according to opportunities we had during the study. As we initially have a focus on Cyber-Physical Systems, we experimented with use cases based on ROS (Robot Operating System), DDS (Data Distribution Service), Redis, Industrial Systems, etc...

At the end of this exploration, a selection was made of the models and use cases kept for comparison. We will present our choices made in order to extract as much useful information as possible from this comparison. This phase revealed a long and arduous work were each use case was combined with each LLMs to generate 5 scripts. Each script was tested on real use cases and analysed. This phase provided interesting conclusions on the use of online or local LLMs but also on the possible use of this kind of tool.

Finally, we will introduce our future work around this topic. We are currenly thinking of how that could extend other tools such as our penetration testing platform (PRACTIS) to fasten the development of reusable penetration test scripts.