央视财经频道《经济半小时》报道度小满用技术破解AI换脸诈骗难题

Core Insights - The report focuses on the financial risks arising from the misuse of AI face-swapping technology, particularly in online lending systems, where criminals attempt to use synthetic facial information for fraud [1][2] - The financial technology company Du Xiaoman has developed specialized risk identification algorithms to combat AI face-swap fraud, providing a critical reference for the industry [1] Group 1: Company Initiatives - Du Xiaoman has created a risk identification algorithm that achieved a risk score exceeding 90 for a live authentication video, indicating a high likelihood of AI face-swap fraud [1] - The company demonstrated its advanced detection model, which can distinguish between real and synthetic faces, with a risk score threshold of 50; all synthetic faces tested scored above this threshold [1] - Du Xiaoman's Chief Technology Officer stated that the detection model is continuously upgraded to keep pace with evolving fraud techniques [1] Group 2: Industry Collaboration - To enhance technical defense capabilities, Du Xiaoman employs a "red-blue confrontation" model for training, simulating attacks and defenses against AI face-swapping [2] - The model's parameters have increased to a hundred billion level, significantly improving its capabilities compared to previous models that had only tens of millions of parameters [2] - In the past year, Du Xiaoman issued fraud alerts to 140,000 customers, successfully preventing over 3,000 fraud cases and potentially saving 180 million yuan in economic losses [2] - Industry experts recommend that financial institutions share information and technology to collectively combat AI face-swap fraud, ensuring timely updates on new fraud techniques [2]