被“误伤”后，他们期待：AI能否让反诈更精准？

Core Viewpoint - The rise of digital fraud has become a significant issue, with the national anti-fraud system actively working to combat it through various measures, although some users experience "collateral damage" due to the system's stringent controls [1][5][9]. Group 1: Anti-Fraud Measures - In 2023, the National Anti-Fraud Center issued 9.4 million funding warning directives, and police departments intervened with 13.89 million individuals, intercepting 2.75 billion fraudulent calls and 2.28 billion fraudulent messages [1]. - The anti-fraud system has intercepted 836.4 million fraudulent domain names and blocked 328.8 billion yuan in suspicious funds [1]. - The implementation of the Anti-Telecom Network Fraud Law in December 2022 has imposed strict obligations on telecom operators and financial institutions to monitor and identify suspicious activities [5]. Group 2: User Experiences - Users have reported sudden suspensions of their phone numbers due to being flagged by the anti-fraud system, often without clear explanations [2][4]. - One user experienced a new card being suspended within a week of activation, leading to confusion and frustration over the lack of clarity regarding the reasons for the suspension [2]. - Another user faced repeated suspensions of her number after making multiple calls, indicating potential flaws in the risk assessment process [4]. Group 3: System Limitations and Improvements - Experts suggest that the anti-fraud system's sensitivity may lead to "collateral damage," where legitimate users are mistakenly flagged as high-risk due to their behavior [9][13]. - The current models primarily analyze superficial data such as call frequency and duration, lacking the ability to understand user intent, which could lead to misclassification of normal activities as suspicious [13][14]. - Recommendations for improvement include upgrading the risk assessment models to incorporate user intent recognition and establishing a more nuanced user profile system to reduce false positives [13][14].