度小满“防深伪”技术揭秘：我们是如何从一段视频里揪出AI造假蛛丝马迹的？

Core Viewpoint - The rise of AI-driven deepfake technology has made it increasingly easy for criminals to create high-quality counterfeit content, posing significant challenges for security in the digital world. Financial technology is leveraging advanced methods to combat these threats and protect users' safety. Group 1: Deepfake Technology and Its Implications - Deepfake technology has become readily accessible, enabling criminals to produce convincing fake content, with instances of scams resulting in significant financial losses, such as a case where 4.3 million yuan was stolen in just 10 minutes using AI face-swapping technology [1]. - The key to identifying AI-generated fakes often lies in subtle flaws that are overlooked by the naked eye, such as abnormal blinking rates, irregular pupil shapes, and defects in teeth rendering [1]. Group 2: Anti-Fraud Initiatives - Chongqing's anti-fraud center and Du Xiaoman have launched a series of AI anti-fraud promotional activities for 2026, including short dramas, a digital anti-fraud song, and an H5 mini-game designed to educate users on recognizing scams through engaging scenarios [1]. - The "Jian Zhen" mini-game simulates real scam scenarios, helping users learn identification techniques in a fun way [1]. Group 3: Advanced Detection Technologies - Du Xiaoman's deep detection technology can uncover hidden "digital fingerprints" in images, allowing for the identification of high-risk behaviors such as impersonation or unauthorized actions [2]. - The technology analyzes unique noise patterns from cameras and AI generators, enabling precise detection of fraudulent activities [2]. Group 4: Micro-Expression Monitoring - The latest advancements in Du Xiaoman's technology include dynamic capture of micro-expressions, which last only about 0.1 seconds and are difficult for the human eye to detect [3]. - The micro-expression risk control model can quantify subtle facial movements, providing insights into the risk level during critical interactions, achieving over 90% recall rate and more than 99% accuracy with a false positive rate of one in a thousand [3]. Group 5: Impact and Future Directions - Du Xiaoman has successfully issued precise fraud warnings to over 450,000 customers, intercepting fraudulent amounts totaling 217 million yuan by leveraging its continuously upgraded anti-deepfake models [3]. - The evolution of technology has not only introduced new scam methods but has also led to the development of robust digital defenses, emphasizing the importance of technology-based identification in fraud prevention [3].