Gan-based Cryptosystem for High Level Data Protection

**GAN-Based Cryptosystem for High Level Data Protection**:

- Réf **ABG-131311**
- Sujet de Thèse
- 19/04/2025
- Contrat doctoral
- Université Bourgogne Europe
- Lieu de travail- Dijon - Bourgogne-Franche-Comté - France
- Intitulé du sujet- GAN-Based Cryptosystem for High Level Data Protection
- Champs scientifiques- Electronique
- Informatique
- Mots clés- Cryptosystem, Generative Adversarial Networks (GANs), Deep-fakes, Images, FPGA, Python, C++

**Description du sujet**:
**Abstract**:
Detecting and safeguarding against malicious deep-fakes while ensuring means to protect private data through secure cryptographic algorithms have become very challenging tasks since the advent of AI in the field of cybersecurity. The aim of this doctoral thesis project is to propose relevant solutions to improve and strengthen the security of digital data, whether it be text, images, or videos by combining the fields of AI/deep learning, cryptography, cryptanalysis, chaotic systems, and FPGA programmable circuits. The approach proposed here is subdivided into two main steps:
1) Utilizing advanced artificial intelligence and machine learning techniques to detect deep-fakes in images or videos that may be manipulated using advanced techniques such as AI. Initially, the idea would be to generate datasets of fake images with high resolution. Then, we use FPGA devices to implement, accelerate, and ensure real-time processing.

2) Employing cryptography and cryptanalysis methods to enhance data protection. This work will involve designing attack techniques based on GANs (AI attacks) to potentially highlight security weaknesses in the existing systems. Finally, designing GAN countermeasures to make the security of the proposed systems robust against attacks. The thesis work will be oriented to FPGA-based design and implementation.

**Project description**:
The integration of AI technologies within cybersecurity frameworks has revolutionized the approach to safeguarding digital assets, enabling proactive threat detection, behavioral analysis, and predictive analytics to anticipate and counter potential attacks before they manifest.

However, while the integration of AI in cybersecurity has heralded significant advancements in threat detection and response capabilities, it has also engendered new challenges and vulnerabilities. One such challenge arises from the malicious exploitation of AI technologies by threat actors to orchestrate sophisticated cyber-attacks, circumvent traditional defense mechanisms, and perpetrate acts of cyber espionage, sabotage, and fraud [Chen 2019, Liao 2021]. The emergence of deep-fake technology exemplifies this trend, wherein adversaries employ AI algorithms to generate hyper-realistic counterfeit content, including forged audio recordings, images, and videos, for deceptive purposes such as disinformation campaigns, social engineering, and identity theft. In addition to the proliferation of deep-fakes, the convergence of AI and cryptography has introduced new dimensions of complexity and vulnerability to cybersecurity ecosystems [Zhu2020, Gomez 2018]. While AI-powered cryptographic solutions hold promise for enhancing data security and privacy, they also present new avenues for adversarial exploitation and cryptographic attacks [Pan 2023]. Adversaries have leveraged AI algorithms to enhance the efficiency and efficacy of cryptanalysis techniques, enabling them to circumvent encryption protocols, decipher sensitive information, and compromise encrypted communications channels. Generative Adversarial Networks (GANs) have emerged as particularly potent tools for cryptanalysis, demonstrating the capability to breach cryptographic defenses and extract confidential information from encrypted data streams [Liu 2021, Lin 2022]. To address the multifaceted challenges posed by the integration of AI in cybersecurity, researchers are exploring innovative approaches to bolster digital defenses and enhance the resilience of critical infrastructure against emerging cyber threats [Min 2024, P. Singh 2024, M. Singh]. One such approach involves the utilization of Field-Programmable Gate Arrays (FPGAs) as hardware-accelerated platforms for implementing AI-driven cybersecurity solutions. FPGAs offer unparalleled flexibility, performance, and energy efficiency, making them well-suited for real-time threat detection, cryptographic operations, and data processing tasks. The integration of AI, cryptography, and FPGA technology represents a paradigm shift in the field of cybersecurity, offering a potent arsenal of tools and techniques for defending against emerging cyber threats and safeguarding the integrity, confidentiality, and availability of digital assets. The expected doctoral thesis work can be divided into 4 main tasks:
1) Design and train a GAN-based cryptosystem to cryptanalyses the standard cryptographic algorithms (AES, ZUC, Chaotic systems, etc.) to detect existing weaknesses facing the new IA-based cybersecurity