Thales, a global leader in aerospace, defense, and digital security, has unveiled an innovative deepfake detection metamodel designed to address the growing threats posed by AI-generated images.
This development comes at a time when malicious actors are increasingly leveraging advanced generative AI platforms, including DALL-E, MidJourney, and Firefly, to create realistic fake content for disinformation campaigns and identity fraud.
According to a statement on Friday, the metamodel, which aggregates various detection techniques, was developed at cortAIx, Thales’s artificial intelligence accelerator.
“Thales’s metamodel employs advanced techniques like Contrastive Language-Image Pre-training to identify mismatches between images and text.
“It also uses the Diffusion Noise Fingerprint method to detect latent noise in AI-generated content and Discrete Cosine Transform analysis to uncover frequency inconsistencies.
“Together, these methods create a powerful framework for detecting deep fake imagery and safeguarding biometric security systems.”
Thales’s Senior Expert in AI and Chief Technology Officer at cortAIx, Christophe Meyer, explained, “The metamodel aggregates multiple detection approaches, including neural networks, noise analysis, and spatial frequency techniques.
“This makes it particularly effective against AI-generated deepfakes, especially in applications requiring biometric identity verification.”
Thales’s metamodel employs advanced techniques like contrastive language-image pre-training to identify mismatches between images and text.
A recent Gartner report estimated that 20 per cent of cyberattacks in 2023 involved deep fake technologies, underscoring the urgency for such innovative solutions.
By targeting the advanced methodologies employed in creating fake images, Thales’s metamodel enhances cybersecurity protocols, particularly in sectors like defense and identity management.
The announcement was made during European Cyber Week, held in Rennes, France, from November 19 to 21.
At the event, Thales demonstrated how the metamodel could be integrated into existing systems to combat identity fraud and improve overall digital security.
The Group highlighted its broader commitment to innovation, supported by an annual R&D investment of €4bn, focusing on artificial intelligence, quantum computing, and cybersecurity.
