CoaP (09/12/2025)
As part of our series La Cybersécurité sur un plateau (Cybersecurity on a Plate), on Tuesday December 9th, we will have two seminars in Room 3.A213 at 10.30 am.
If you are coming to participate for the first time, do not hesitate to contact the organizers so as not to be blocked at the entrance.- Virgile Prevosto (CEA LIST): Preuve d'intégrité avec Frama-C/MetAcsl
Abstract: Frama-C est un outil d'analyse de programmes C, muni d'un langage de spécifications formelles ACS, dans lequel l'utilisateur peut décrire les propriétés attendues du code, principalement sous la forme de contrats de fonction et d'assertions logiques. Ce type d'annotations est bien adapté pour décrire des propriétés fonctionnelles du programme, c'est à dire intuitivement que le programme calcule bien ce qu'on attend de lui. En revanche, d'autres classes de propriétés sont difficilement exprimables en ACSL, exigeant un encodage complexe et donc source potentielle d'erreur dans la spécification elle-même. C'est typiquement le cas pour des propriétés de sécurité comme l'intégrité ou la confidentialité du contenu de locations mémoire du programme, qui demanderaient en théorie d'ajouter des assertions pour chaque accès en écriture (respectivement en lecture), ce qui devient rapidement rédhibitoire, même sur un code de taille modest
Le greffon MetAcsl de Frama-C a été développé par Virgile Robles dans le cadre de sa thèse pour répondre à ce problème. Plus précisément, MetAcsl permet de définir dans un langage dédié des HILAREs (High-Level ACSL Requirements), propriétés destinées à être instanciées automatiquement en annotations ACSL à chaque point du programme correspondant au contexte de l'HILARE (par exemple chaque accès en écriture/en lecture). En outre, MetAcsl permet aussi dans certains cas de raisonner directement au niveau des HILAREs. Une telle étape de "méta-déduction" permet de valider une HILARE en fonction de celles précédemment prouvées, évitant ainsi de devoir prouver séparément toutes ses instances au niveau ACSL
Dans cet exposé, nous présenterons les principales constructions des HILAREs ainsi que les mécanismes de validation, par preuve directe des annotations ACSL générées via le greffon WP et par méta-déduction au sein de MetAcsl, à travers une étude de cas visant à établir l'intégrité de l'automate qui supervise le déroulement du bootloader de WooKey, un prototype de clé USB sécurisée développée par l'ANSSI.
Bio: Virgile Prevosto est Ingénieur Chercheur au laboratoire de sûreté et sécurité des logiciels du CEA List, expert senior en méthodes formelles. Il est un des principaux développeurs de la plateforme d'analyse de programmes Frama-C.
- Benoit NOUGNANKE: How Dataset Diversity Affects Generalization in ML-based NIDS
Abstract: Machine Learning-based Network Intrusion Detection Systems (ML-based NIDS) rely heavily on the quality of the datasets used for training and evaluation. However, widely used NIDS benchmarks often suffer from poor data diversity, which limits model generalization and undermines the reliability of evaluation protocols. While prior work has acknowledged this limitation, a systematic framework to quantify dataset diversity and analyze its relationship with performance is still missing. To address this gap, we introduce a structured approach for characterizing dataset diversity in ML-based NIDS, grounded in measurement theory. We distinguish three types of diversity—intra-class, inter-class, and domain-shift—and operationalize their measurement using established metrics such as the Vendi Score and the Jensen-Shannon divergence. Our empirical analysis on the CIC-IDS2018 dataset, spanning sixty diversity-controlled train–test experiments, provides new insights into the relationship between diversity and generalization and demonstrates the value of diversity-aware data sampling for improving evaluation reliability.
- Virgile Prevosto (CEA LIST): Preuve d'intégrité avec Frama-C/MetAcsl
Students C4 seminar (23/10/2025)
The seminar will start 2 pm at IMT (Amphi 2). It will be followed by a cocktail at the Entrepotes 19 restaurant.
-
Agenda
- 1.30 pm: Welcome Coffee
- 2 pm: Grégory Blanc, Christophe Kiennert, Olivier Levillain -Opening
- 2.15 pm: Gabriel Zaïd (Thales) - Vers une nécessité de Certifier la sécurité des IA - Exemple d’une attaque physique
- 3 pm: Karolina Gorna (Télécom Paris) - La détection de vulnérabilités par exécution concolique dans les binaires Go
- 3.30 pm: Nicolas Peiffer (Thales) - Key Management Service plugin pour Kubernetes
- 4 pm: Break
- 16.45 pm: Grégoire Menguy (CEA) - Synthèse de programme pour la rétro-ingénierie de binaire
- 5.30 pm: Aina Rasoamanana (Valéo) - Utilisation de l'Active Automata Learning pour analyser les implémentations des protocoles réseau
- 6 pm: Rump Session
- about 7 pm: dinner cocktail at the 19.
Gabriel Zaïd (Thales) - Vers une nécessité de Certifier la sécurité des IA - Exemple d’une attaque physique
La diffusion croissante des systèmes d’intelligence artificielle (IA) dans des domaines critiques, tels que celui de la défense, impose une réflexion sur leur certification en matière de sécurité. Le règlement européen AI Act répond à cet impératif en instaurant un cadre réglementaire destiné à évaluer et contrôler les risques liés aux IA dites « à haut risque ». L’objectif est double : garantir la confiance des utilisateurs et prévenir les dérives techniques ou éthiques associées à ces technologies. À travers cette présentation, nous montrerons, par l’exemple, la mise en œuvre d’une attaque contre une IA embarquée dans un système physique. Cette attaque, combinant des attaques par canaux auxiliaires et par cryptanalyse, vise à porter atteinte à la propriété intellectuelle d’une entreprise en copiant fidèlement l’IA ciblée. Nous expliquerons en quoi une telle menace peut compromettre la confiance dans les systèmes d’IA et présenterons quelques pistes de réflexion permettant de réduire ces risques.
Bio : Gabriel ZAÏD est ingénieur en cryptographie et machine learning au CESTI de Thales à Toulouse. Il évalue la sécurité des instruments physiques et des systèmes embarquant des primitives cryptographiques. Ses travaux de recherche couvrent plusieurs aspects pratiques autour de la cryptographie et son application dans les sytèmes embarqués. Il s'intéresse particulièrement aux attaques par canaux auxiliaires et à l'usage du machine learning, notamment afin d'acquérir une meilleure compréhension de la sécurité physique de l'intelligence artificielle embarquée.
Karolina Gorna (Télécom Paris) - La détection de vulnérabilités par exécution concolique dans les binaires Go
Go est devenu essentiel pour les infrastructures cloud et les applications blockchain, mais ses programmes restent vulnérables à des défaillances d'exécution que les tests conventionnels ne détectent souvent pas. Les outils d'exécution symbolique existants peinent à gérer les fonctionnalités spécifiques au runtime de Go, notamment l'ordonnancement des goroutines, la gestion de la mémoire et la répartition des interfaces. Notre travail présente Zorya, un framework d'exécution concolique implémenté en Rust qui opère sur des binaires pouvant être traduits dans la représentation intermédiaire P-Code bas niveau de Ghidra, dont notamment les binaires Go. Le framework introduit des techniques d'exploration guidées par les panics qui concentrent l'effort d'analyse sur les chemins de code critiques, combinées à des stratégies d'analyse au niveau des fonctions. L'évaluation démontre que Zorya détecte avec succès diverses catégories de vulnérabilités d'exécution dans les programmes Go, incluant des cas de test théoriques et réels issus d'audits de sécurité.
Bio : Karolina GORNA est doctorante en cybersécurité et blockchain à Télécom Paris et au Ledger Donjon. Ancienne diplômée du master SSR de Télécom SudParis et présidente de l'association KRYPTOSPHERE, elle a encadré plus de 500 étudiants à travers la France et co-organisé l'International Space Apps Challenge de la NASA à Paris. Elle a mené des formations pour le MIT Professional Education, l'AFORP et Télécom Paris, et participe actuellement aux travaux du Campus Cyber sur la sécurité des cryptoactifs.
Nicolas Peiffer (Thales) - Key Management Service plugin pour Kubernetes
Le k8s-kms-plugin est un plugin de Key Management Service (KMS v2 Provider) pour Kubernetes. Il se destine principalement aux cluster kubernetes dit edge ou far edge, qui ont la particularité d'être contraint en ressource (comme des IoT) et contraints en terme de connectivité (connexion à un réseau internet intermittente ou inexistante). Il utilise des racines de confiance matérielles comme des TPM ou des USB HSM pour y stocker des Key Encryption Key qui servent à protéger en confidentialité (chiffrement) une Data Encryption Key utilisée par l'API Kubernetes pour chiffrer des élements sensibles comme les objets "secrets" Le plugin est compatible avec des équipements PKCS #11 comme les Thales SafeNet eToken Fusion (2)(3) ou les Yubico YubiHSM. Codé en Go, il offre une CLI moderne qui facilite son utilisation.
Bio : Diplômé de Télécom SudParis en 2017, titulaire du tire ESSI, je me suis forgé une expérience avec les technologies et les paradigmes du monde "cloud-natif" appliqués aux systèmes industriels critiques. Openstack, QEMU, Kubernetes, Podman, DevSecOps, ingénierie logicielle, Supply chain security, architecture cybersécurité, cryptographie post quantique... sont quelques mots clés pour décrire les sujets que je traite au quotidien. Je suis un humble contributeur sur plusieurs projets open source, comme par exemple sur la spécification SBOM CycloneDX. Je suis également un des points de contacts entre Thales et la Linux Foundation.
Grégoire Menguy (CEA) - Synthèse de programme pour la rétro-ingénierie de binaire
La rétro-ingénierie de programme, et notamment de binaires, est une tâche cruciale en sécurité, par exemple pour comprendre des logiciels malveillants. Malheureusement, les logiciels sont de plus en plus grands et complexes. Il devient donc nécessaire de proposer de nouvelles approches automatiques pour aider à la rétro-ingénierie et à la compréhension de code. Celles-ci sont usuellement en boîte blanche, utilisant la syntaxe du code pour déduire ses propriétés. Elles sont très efficaces mais sont impactées par la complexité syntaxique du code qui peut être accentuée par de l'obfuscation. Cette présentation explore comment les méthodes en boîte noire peuvent inférer des propriétés utiles pour la rétro-ingénierie. Nous étudierons deux problèmes : (i) l'inférence de contrat de fonction qui tente d'apprendre sur quelles entrées une fonction peut être exécutée pour obtenir les sorties souhaitées et (ii) la déobfuscation, qui vise à simplifier du code obfusqué.
Bio : Grégoire Menguy est chercheur au CEA LIST. Ses recherches se concentrent sur l’utilisation des méthodes d’intelligence artificielle pour la rétro-ingénierie et la déobfuscation. Il a réalisé sa thèse au CEA LIST après un cursus d'ingénieur à Telecom SudParis (option cybersécurité) où il a été certifié ESSI par l'ANSSI.
Aina Rasoamanana (Valéo) - Utilisation de l'Active Automata Learning pour analyser les implémentations des protocoles réseau
Les implémentations de protocoles réseau sont omniprésentes dans nos systèmes modernes. Nous nous appuyons quotidiennement sur divers protocoles tel que TLS. L'un des problèmes des piles réseau est qu'elles peuvent présenter des transitions incorrectes dans leurs machines à état, ce qui peut entraîner des problèmes de sécurité. Dans un précédent artic, notre équipe a étudié les machines à état de diverses piles TLS, ce qui nous a permis de rejouer des bugs de sécurité connus et de découvrir des nouvelles vulnérabilités. Globalement, nos découvertes peuvent être classées comme suit :
- des contournements d'authentification (en sautant un ou plusieurs messages, il peut être possible d'atteindre des états authentifiés sans avoir à présenter des signatures ou mot de passe) ;
- des dénis de service (dans certains cas, nous avons trouvé des états où une pile accepterait indéfiniment un message insignifiant, permettant à un attaquant de maintenir une connexion ouverte, avant l'authentification) ;
- le fingerprinting (ce n'est pas une vulnérabilité en soi, mais nous avons montré que les piles TLS pouvaient être distinguées par leur comportement).
Depuis 2022, nous travaillons sur d'autres protocoles tels que SSH, et sur l'amélioration de nos outils. Cette étude a soulevé des défis intéressants en termes de complexité (par exemple, SSH utilise beaucoup plus de messages par rapport à TLS et peut produire d'énormes machines à état) et d'expressivité (exprimer et raisonner sur les propriétés de sécurité de manière rigoureuse et efficace peut être difficile).
Bio : Aina Rasoamanana est ingénieur R&D en sécurité à Valeo. Titulaire d’un doctorat de l’Institut Polytechnique de Paris, réalisé à Télécom SudParis, son expertise s’étend à la cryptographie, à la sécurité des protocoles de communication et à la sécurité logicielle.
CoaP (09/09/2025)
As part of our series La Cybersécurité sur un plateau
(Cybersecurity on a Plate), on Tuesday September 9th, we
will have two seminars in Room 3.A213 at 10.30 am.
If you are coming to participate for the first time, do not
hesitate to contact the organizers so as not to be blocked at the
entrance.
- Guillaume Scerri (ENS Paris-Saclay): Capturing new cryptographic proofs techniques using logic: extensions to the computationally complete symbolic attacker
Abstract: When proving cryptographic protocols, one has to deal with a malicious attacker. In particular this means that it is crucial to carefully capture attacker capabilities. This is best done by reducing security of protocols to known hard problems. However these cryptographic reductions can be quite complex and when done by hand such reductions can be rather hard to check. In recent years there has been a push to capture such reductions using logics that can be checked using (dedicated) proof assistants. In this talk we explore the intricacies of capturing complex cryptographic reductions in one of these logics the Computationally Complete Symbolic Attacker logic. We will focus on two main techniques: hybrid arguments and rewinding, and show how they can be applied to proofs of new protocols, namely e-voting mixnets.
- Vidal Attias (CEA): Augmenting Search-based Program Synthesis with Local Inference Rules to Improve Black-box Deobfuscation
Abstract: Code obfuscation aims to protect programs from reverse engineering, with applications ranging from intellectual property protection to malware hardening. Recent works on black-box analyses propose to leverage program synthesis in order to infer the semantics of highly obfuscated code blocks. Being fully black-box, these approaches are immune to syntactic complexity and can thus bypass standard obfuscation mechanisms. Yet, they are restricted by their synthesis capabilities and can only be applied to semantically simple code blocks. It explains why they have mainly been used on virtual machine handlers, where behaviors are usually simple enough. Applying black-box deobfuscation at scale beyond virtualization is still an open problem, notably because black-box methods cannot synthesize complex behaviors involving, for example, arbitrary constant values or affine or polynomial relations over mixed-boolean-arithmetic expressions. In this article, we show how to combine search-based program synthesis with local inference rules, resulting in a black-box method named Search Modulo Inference Rules (Smir) which allows boosting the capabilities of search-based program synthesis while keeping its generality and flexibility.
We instantiate our method with inference rules dedicated to hard synthesis problems like arbitrary constant values and affine or polynomial relations over mixed boolean expressions, yielding a new black-box deobfuscation tool named XSmir. Experiments demonstrate that XSmir significantly outperforms prior black-box deobfuscators.
CoaP (10/06/2025)
As part of our series La Cybersécurité sur un plateau
(Cybersecurity on a Plate), on Tuesday June 10th, we
will have one seminar in Room 3.A213 at 10.30 am.
If you are coming to participate for the first time, do not
hesitate to contact the organizers so as not to be blocked at the
entrance.
- Héliou Alice, Thouvenot Vincent, Lampe Rodolphe, Huynh Cong Bang, Morisse Baptiste (Thales): AI Friendly Hacker : when an AI reveals more than it should
Abstract: The aim of AI based on machine learning is to generalize information about individuals to an entire population. And yet...
- Can an AI leak information about its training data?
- Since the answer to the first question is yes, what kind of information can it leak?
- How can it be attacked to retrieve this information?
To emphasize AI vulnerability issues, a challenge was proposed at CAID2023 on confidentiality attacks based on two tasks:
- Membership Attack: An image classification model has been trained on part of the FGVC-Aircraft open-access dataset. The aim of this challenge is to find, from a set of 1,600 images, those used for training the model and those used for testing.
- Forgetting attack: The model supplied, also known as the export model, was refined from a so-called sovereign model. The sovereign model has certain sensitive aircraft classes (families) which have been removed and replaced by new classes. The aim is to find which of a given set of classes have been used to train the sovereign model, using only the weights of the export model.
The Friendly Hackers team of CortAIx LAbs won the two tasks. At the seminar we will present how we did it and what lessons we learned during this fascinating challenge.
CoaP (13/05/2025)
As part of our series La Cybersécurité sur un plateau
(Cybersecurity on a Plate), on Tuesday May 13th, we
will have two seminars in Room 3.A213 at 10.30 am.
If you are coming to participate for the first time, do not
hesitate to contact the organizers so as not to be blocked at the
entrance.
- Dimitrios Kokkonis (CEA) - ROSA: Finding Backdoors with Fuzzing
Abstract: A code-level backdoor is a hidden access, programmed and concealed within the code of a program. For instance, hard-coded credentials planted in the code of a file server application would enable maliciously logging into all deployed instances of this application. Confirmed software supply-chain attacks have led to the injection of backdoors into popular open-source projects, and backdoors have been discovered in various router firmware. Manual code auditing for backdoors is challenging and existing semi-automated approaches can only handle a limited scope of programs and backdoors, while requiring manual reverse-engineering of the audited (binary) program. Graybox fuzzing (automated semi-randomized testing) has grown in popularity due to its success in discovering vulnerabilities and hence stands as a strong candidate for improved backdoor detection. However, current fuzzing knowledge does not offer any means to detect the triggering of a backdoor at runtime. In this work we introduce ROSA, a novel approach (and tool) which combines a state-of-the-art fuzzer (AFL++) with a new metamorphic test oracle, capable of detecting runtime backdoor triggers. To facilitate the evaluation of ROSA, we have created ROSARUM, the first openly available benchmark for assessing the detection of various backdoors in diverse programs. Experimental evaluation shows that ROSA has a level of robustness, speed and automation similar to classical fuzzing. It finds all 17 authentic or synthetic backdooors from ROSARUM in 1h30 on average. Compared to existing detection tools, it can handle a diversity of backdoors and programs and it does not rely on manual reverse-engineering of the fuzzed binary code.
Bio: I am a PhD student in the BINSEC team at CEA List, working under the supervision of Stefano Zacchiroli and Michaël Marcozzi. My research is focused on the automation of the detection of advanced vulnerabilities in binary programs. I graduated from Polytech Sorbonne in 2020 with a Master's degree in Embedded Systems.
- Quentin Michaud (Télécom SudParis / Thales) - Robust Stack Smashing Protection for WebAssembly
Abstract: WebAssembly is an instruction set architecture and binary format standard, designed for secure execution by an interpreter. Previous work has shown that WebAssembly is vulnerable to buffer overflow due to the lack of effective protection mechanisms. In this work, we evaluate the implementation of Stack Smashing Protection (SSP) in WebAssembly standalone runtimes, and uncover two weaknesses in their current implementation. The first one is the possibility to overwrite the SSP reference value because of the contiguous memory zones inside a WebAssembly process. The second comes from the reliance of WebAssembly on the runtime to provide randomness in order to initialize the SSP reference value, which impacts the robustness of the solution. We address these two flaws by hardening the SSP implementation in terms of storage and random generator failure, in a way that is generalizable to all of WebAssembly. We evaluate our new, more robust, solution to prove that the implemented improvements do not reduce the efficiency of SSP.
Bio: I am a Télécom SudParis PhD student in the Cybersecurity team of CortAIx Labs, a Thales research laboratory, under supervision of Joaquin Garcia-Alfaro, Olivier Levillain and Dhouha Ayed. I am working on securing distributed systems on constrained and diverse devices by leveraging technologies such as WebAssembly and Confidential Computing.
CoaP (08/04/2025)
As part of our series La Cybersécurité sur un plateau
(Cybersecurity on a Plate), on Tuesday April 8th, we
will have two seminars in Room 3.A213 at 10.30 am.
If you are coming to participate for the first time, do not
hesitate to contact the organizers so as not to be blocked at the
entrance.
- Ayoub Wehby (Telecom Paris) - Towards Secure Connected Cars: AI-Based Defense Against CAM-Based DDoS Attacks
Abstract: The increasing connectivity of modern cars enhances driver safety and comfort but also expands the attack surface for cyber threats. In this presentation, we first explore the vulnerabilities of connected cars, focusing on Distributed Denial-of-Service (DDoS) attacks leveraging Cooperative Awareness Messages (CAMs) and their impact on safety-critical applications. We then introduce a machine-learning detection approach, developed using a CAM-based DDoS dataset generated from a realistic traffic scenario in Luxembourg City. Next, we demonstrate the generalizability of our models against morphing DDoS attacks. Finally, we unveil a new attack model incorporating Sybil-based techniques that challenge our detection system and discuss the strategies employed to restore detection accuracy. This work highlights the urgent need for robust intrusion detection systems in connected car environments.
- Renaud Sirdey (CEA) - Beyond CPA security for FHE
Abstract: Since its inception more than ten years ago, Fully Homomorphic Encryption has been the subject of a lot of research towards more efficiency and better practicality. From a security perspective, however, FHE still raises a number of questions and challenges, in particular due to the fact that all the FHE used in practice achieve only CPA-security (and all of these schemes are trivially CCA1 insecure). Over the last few years, very active research has been done to explore the security of FHE beyond that regime with new security notions, attacks and constructions emerging. In this talk, we will cover recent “slightly beyond CPA” security notions, such as CPAD, as well as new attacks on FHE in that model. We’ll then move on to CCA security for FHE and present recent results towards answering the two questions: can we build FHE schemes offering some degree of CCA security? And, what is the strongest degree of CCA-security achievable by FHE?
LINCS seminar (19/03/2025)
As part of our collaboration with LINCS, on Wednesday March the 19th, we will have the following seminar (Palaiseau, Amphi 3, 2 pm):
- Romain Dagnas - IRT SystemX & Telecom SudParis
Abstract: Thanks to technological advancements, critical infrastructures integrate many smart technologies and become highly connected to the cyber world. This is especially true for Cyber-Physical Systems (CPSs), which combine hardware and software components. Despite the advantages of smart infrastructures, they remain vulnerable to cyber threats and adversarial events such as cyber-attacks. This talk focuses on quantifying the cyber resilience of complex systems modeled with a multi-layering approach. As a use case, we consider the Secure Water Treatment System (SWaT) testbed.
[Recording]
CoaP (11/03/2025)
As part of our series La Cybersécurité sur un plateau
(Cybersecurity on a Plate), on Tuesday March 11th, we
will have two seminars in Room 3.A213 at 10.30 am.
If you are coming to participate for the first time, do not
hesitate to contact the organizers so as not to be blocked at the
entrance.
- Nicolas Peiffer (Thales) - A Journey Through SBOMs and Software Provenance Attestations in the Industry: Cryptography BOM, Patents BOM, ML-AI BOM, Meta-BOM...
Abstract: In Europe, the Cyber Resilience Act (EU CRA) is being implemented to encourage, through regulations and laws, companies and open-source software communities to develop more secure software. The EU CRA is often referred to as the "GDPR for software": although the directive is now in effect, many entities are not yet prepared and face technical and organizational questions that they will need to address in order to comply with the legislation. This presentation will share Thales' experience regarding Software Bills of Materials (SBOMs) and software provenance attestations, such as in-toto and SLSA. It will particularly focus on "exotic BOMs," including Cryptography BOM, Patents BOM, ML-AI BOM, and Meta-aggregated-BOM, for which there are few or no suitable tools available. The presentation will also discuss the challenges associated with the Meta-aggregated-BOM in the context of "system of systems." Finally, it will highlight Thales' open-source contributions to the CycloneDX BOM format.
- Guilhem Lacombe (CEA) - Attacker Control and Bug Prioritization
Abstract: As bug-finding methods improve, bug-fixing capabilities are exceeded, resulting in an accumulation of potential vulnerabilities. There is thus a need for efficient and precise bug prioritization based on exploitability. In this work, we explore the notion of control of an attacker over a vulnerability’s parameters, which is an often overlooked factor of exploitability. We show that taint as well as straightforward qualitative and quantitative notions of control are not enough to effectively differentiate vulnerabilities. Instead, we propose to focus analysis on feasible value sets, which we call domains of control, in order to better take into account threat models and expert insight. Our new Shrink and Split algorithm efficiently extracts domains of control from path constraints obtained with symbolic execution and renders them in an easily processed, human-readable form. This in turn allows to automatically compute more complex control metrics, such as weighted Quantitative Control, which factors in the varying threat levels of different values. Experiments show that our method is both efficient and precise. In particular, it is the only one able to distinguish between vulnerabilities such as cve-2019-14192 and cve-2022-30552, while revealing a mistake in the human evaluation of cve-2022-30790. The high degree of automation of our tool also brings us closer to a fully-automated evaluation pipeline.
CoaP (11/02/2025)
As part of our series La Cybersécurité sur un plateau
(Cybersecurity on a Plate), on Tuesday February 11th, we
will have two seminars in Room 3.A213 at 10.30 am.
If you are coming to participate for the first time, do not
hesitate to contact the organizers so as not to be blocked at the
entrance.
- Adam Oumar Abdel-rahman (Télécom SudParis) - Privacy-Preserving Web Content Filtering using Attribute-Based Encryption
Abstract: The rapid growth of encrypted data, particularly TLS-encrypted traffic, has enhanced privacy by reducing cyberattack exposure. However, this shift challenges traditional monitoring tools, rendering them ineffective in handling encrypted traffic. Balancing confidentiality and security in systems such as intrusion detection requires innovative solutions. In this paper, we explore privacy-preserving web content filtering using attribute-based encryption (ABE). This approach enables enforcing access policies—such as blocking inappropriate or harmful content—while preserving users' privacy and requests. We present a comprehensive study, from specification to evaluation, showcasing how cryptographic techniques can address the dual needs of privacy and control in modern information systems.
- Julien Malka (Télécom Paris) - Increasing trust in the open source software supply chain with reproducible builds and functional package management
Abstract: Functional package managers and reproducible builds are technologies and methodologies that are conceptually very different from the traditional software deployment model, and that have promising properties for software supply chain security. In this presentation, I’ll introduce the main lines of work included in my thesis surrending these topics and the findings from a research project on the reproducibility of the Nix software repository.
CoaP (14/01/2025)
As part of our series La Cybersécurité sur un plateau
(Cybersecurity on a Plate), on Tuesday January 14th, we
will have one seminar in Room 3.A213 at 10.30 am.
If you are coming to participate for the first time, do not
hesitate to contact the organizers so as not to be blocked at the
entrance.
- Sara Chennoufi (Télécom SudParis) - Towards Interpretable and Resilient Cyber Intrusion Detection in Heterogeneous Environments
Abstract: In distributed networks, devices face diverse cyberattacks, highlighting the need for collaborative mechanisms like Federated Learning (FL) to achieve a global knowledge of various attack types. FL is a privacy-preserving machine learning paradigm that enables collaborative model training without sharing sensitive data. However, the effectiveness of FL is often hindered by the heterogeneity of attack data across different networks known as non-identically distributed (non-IID) data. To address these challenges, we propose Argos, a Federated Prototype Learning (FPL) framework designed to improve collaborative and interpretable detection of network attacks in heterogeneous environments. FPL enables the sharing of class-specific prototypes, facilitating the exchange of interpretable knowledge and improving the detection performance for individual classes. Additionally, we analyze the privacy risks associated with prototype sharing and investigate their effectiveness in identifying mislabeled data.
Biography: Sara Chennoufi is a PhD candidate at Télécom SudParis, working on the development of intrusion detection systems for 5G networks. Her research focuses on enhancing privacy-preserving distributed systems using Federated Learning, addressing key challenges in 5G such as system heterogeneity and the rapid spread of new cyber attacks. She graduated from the Higher School of Computer Science (ESI Algiers). For her graduation project, she completed an internship at INSA Lyon, where she also explored Federated Learning and poisoning attacks.
Seminars in: [2024] [2023] [2022]