Fr

Three Projects Selected for IMC2’s First Call for Projects

IMC2 launched its very first call for projects focused on promoting and protecting the right to privacy in Quebec.

The call generated strong interest, with eight projects submitted, bringing together 22 professors who are members of our assembly, as well as international partners, notably from the Pôle d’Excellence Cyber in France and CyberWal in Wallonia. Three interuniversity projects were selected by the Evaluation Committee, highlighting the richness of expertise and collaboration within the Institute.

All submitted proposals demonstrated a high level of quality and relevance. The three selected projects stood out for their structuring nature and strong impact potential.

Measuring the Invisible: A Benchmark for Evaluating Internal Sensitive Data Leakage in LLM-Based Multi-Agent Systems

Description

Large language model multi-agent systems (MA-LLMs) can break down complex tasks and delegate subtasks to specialized agents. As these systems are increasingly deployed on sensitive data, they raise a critical question: do they process sensitive information securely and confidentially?

Audits of real-world workflows across multiple domains reveal that, even when final outputs comply with privacy policies, sensitive data may still circulate undetected through internal communications between agents. These systems therefore introduce data privacy risks that current evaluation benchmarks fail to measure, as sensitive information can pass through channels that output-focused audits never inspect.

This project aims to design an end-to-end benchmark explicitly covering internal channels in order to quantify data leakage and prevent privacy breaches from remaining undetected.

Team

  • Ranwa Al-Mallah – Associate Professor, Polytechnique Montréal
  • Raphaël Khoury – Professor, Université du Québec en Outaouais
  • Samiha Ayed – Professor, IMT Atlantique, France
  • David Espès – Professor, Université de Bretagne Occidentale, Lab -STICC, France
  • Solutions Flow Factor Inc.
  • Fomox Inc.

The Impact of LLMs on Secure Programming Education: Strategies and Perspectives

Description

The emergence and growing use of LLMs have profoundly transformed several areas of computer science, particularly programming. This new technology has become indispensable for developers, whether beginners or experienced professionals, in code development. However, programming education methods have not kept pace with this evolution and are often outdated.

Several studies have shown that code generated with the help of LLMs frequently lacks adequate security. These same studies, however, point to promising usage strategies that enable the generation of secure code.

The objective of this research project is to develop state-of-the-art programming strategies aligned with the evolution of this technology, along with appropriate teaching methods, in order to introduce students to cybersecurity best practices related to the use of LLMs and better prepare them to meet growing market demand in this field.

Team

  • Raphaël Khoury – Professor, Université du Québec en Outaouais
  • Omar Abdul Wahab – Assistant Professor, Polytechnique Montréal
  • Alan Davoust – Professor, Université du Québec en Outaouais

Graph Models for Differentially Private Data Analysis and Information System Protection

Description

This project aims to strengthen the privacy and security of modern information systems, which are characterized by massive data exchanges and complex interactions. It proposes modeling information systems as graphs to better analyze dependencies, information flows and emerging risks.

Two complementary research directions are explored:

  1. The design of graph data analysis mechanisms offering differential privacy guarantees, particularly for the real-time processing of structured signals;
  2. The study of systemic vulnerabilities in access architectures and inference risks through structural and dynamic models of information propagation.

Expected contributions include new algorithmic and analytical tools, operational risk indicators and information risk mitigation methods. The project will support digital systems governance, personal data protection, scientific outreach and the training of highly qualified personnel (HQP).

Team

  • Jérôme le Ny – Full Professor, Polytechnique Montréal
  • Étienne Gaël Tajeuna – Professor, Université du Québec en Outaouais
  • Shuang Gao– Assistant Professor, Polytechnique Montréal