Pierre-Louis Goudet
Directeur(s): David Brosset
Encadrant(s): Paul Perrotin & Douraïd Naouar

Obfuscation of Human Activities in Encrypted Communications: Application to Maritime Industrial Networks
Ph.D. Context
The protection of data exchanged over digital networks is one of the major societal challenges, whether it concerns online privacy or confidential information processed by the military.
Several mechanisms ensure the security of communications on digital networks. The most crucial is data encryption. A sufficiently strong encryption method ensures that decrypting the data without possessing the decryption key is so improbable that it is considered impossible. Indeed, state-of-the-art encryption algorithms rely on mathematical problems that have been proven to be unsolvable within a reasonable timeframe. The advent of quantum computing has been widely anticipated, and post-quantum algorithms are already available.
However, encryption applies only to the payload of transmitted messages and not to network frames, which must be decoded to travel through network nodes from sender to receiver. Metadata and statistical analysis of these frames can reveal numerous insights that should remain inaccessible.
Several prior works are related to this subject, particularly studies on intent identification in cyberattacks (Merien et al., 2018) and labeled network traffic generation (Nogues et al., 2019). The Ph.D. of Maxence Lannuzel (CIFRE Thesis, Interface Concept, 2022-2025) focuses on identifying digital activities within networks through the design of an intelligent switch (Lannuzel & Brosset, 2024). These studies build upon activity theory, which has previously been used to study intermodal urban mobility in Ines Jguirim's thesis (AER, 2012-2016). The results demonstrate that activity identification is effective on encrypted flows using robust algorithms resilient to false packet injection. A digital activity fingerprinting method is currently under development.
Based on Maxence Lannuzel’s findings, it appears that network communication classification and network analysis are feasible (A. Bozorgi et al., 2023). This threatens individuals' privacy protection and the confidentiality of sensitive information.
Several studies have explored communication obfuscation (W. Li et al., 2023, Y.-W. Lee et al., 2022, Meier et al., 2022), primarily aiming to prevent traffic analysis but without addressing the generated noise that signals modifications to the traffic.
The proposed research aims to mimic real digital activities to camouflage actual activities while minimizing noise. This solution is also beneficial for preserving quality of service. By modifying activities through packet injection based on well-identified activities, it is possible to maintain the original network traffic dimensions.
Scientific Challenges
The key scientific challenges include:
Naval Interest
Ubiquitous communication encryption creates a false sense of security and anonymity. The emergence of increasingly powerful AI-based network communication classification algorithms poses a threat to the protection of sensitive information. Moreover, network traffic categorization can enable network topology recognition, facilitating targeted cyberattacks.
The advanced network obfuscation methods developed in this research will enhance communication protection and improve the security of critical system-of-systems.
References