MEUDON, France – The French Ministry of Defence faced a significant challenge in AI security, and Thales, a global leader in advanced technologies, emerged victorious. Thales’ Friendly Hackers team successfully performed a sovereign AI hack, demonstrating their expertise in protecting sensitive data and trained models.
The challenge, known as the CAID challenge, consisted of two tasks. The first task required participants to determine which images were used to train the AI algorithm and which were used for testing. Thales’ Friendly Hackers team, through a thorough analysis of the AI model, successfully identified the training images, gaining valuable insights into the training methods and model quality.
The second task was to find all the sensitive images of aircraft that had been protected using unlearning techniques. Unlearning involves deleting the data used to train a model to preserve its confidentiality. The goal is to prevent the extraction of sensitive data from the model in case of theft, loss, or export. Thales’ team managed to re-identify the supposedly erased data, bypassing the unlearning process. This exercise highlights the importance of assessing the vulnerability of training data and trained models, as they can be potential targets for malicious attacks.
Attacks on training data or trained models could have catastrophic consequences, especially in a military context. Infiltration of these systems could lead to model theft, extraction of sensitive data used for recognizing military hardware, or the injection of malware and backdoors, compromising the entire AI system.
Recognizing the critical need for AI cybersecurity, Thales has developed a comprehensive set of countermeasures called the BattleBox. These countermeasures aim to enhance the protection of training data and trained models. Encryption of the AI model is one of the solutions Thales suggests to prevent AI hacking.
David Sadek, Thales VP Research, Technology & Innovation in charge of Artificial Intelligence, emphasized the importance of high levels of security and cybersecurity for AI. He stated, Thales implements a large range of AI-based solutions for all types of civil and military use cases and highlighted Thales’ expertise in incorporating AI solutions into their systems to improve their operational capabilities.
Thales has invested significantly in developing technical capabilities to test the security of AI algorithms and neural network architectures. Their Friendly Hackers team at the ThereSIS laboratory achieved first place on both tasks in the AI challenge, showcasing Thales’ expertise in AI security.
As the use of AI in defense and security continues to grow, Thales has developed an ethical and scientific framework for trusted AI. Their solutions combine the expertise of over 300 senior AI experts and more than 4,500 cybersecurity specialists.
Thales, a global leader in advanced technologies, is dedicated to making the world safer, greener, and more inclusive. With a strong focus on research and development, the company invests close to €4 billion annually in key areas such as quantum technologies, Edge computing, 6G, and cybersecurity. Thales employs 77,000 people across 68 countries and generated €17.6 billion in sales in 2022.
The successful outcome of Thales’ sovereign AI hack highlights the critical need for robust AI cybersecurity measures in the defence sector. As AI technologies continue to evolve, the protection of sensitive data and trained models will play a vital role in maintaining security and thwarting potential breaches.
