*ÌÇÐÄlogo¹ÙÍøÊ×Ò³

• China Unicom and ZTE launched the Smart Safeguard, an AI-driven anti-fraud system, recognized for its innovative technology
• The solution improves fraud detection efficiency, achieving over 95% accuracy and reducing manual review costs

Shenzhen, China, 13 November 2024 - Smart Safeguard, an advanced multimodal large language model (MLLM) and multilingual anti-fraud system jointly launched by China Unicom and ZTE, has been awarded the top prize for "2023 Innovative Technology for Telecommunications Anti-Fraud" by the Anti-Fraud Task Force of the Ministry of Industry and Information Technology (MIIT) of China in Q2 2024. This prestigious recognition highlights the system's outstanding performance in combating fraud since its launch in July of the previous year. In October 2023, ZTE officially released the industry's first Smart Safeguard anti-fraud governance system based on AI large models.

According to the 2023 annual report from the Global Anti Scam Alliance (GASA), online scams have caused approximately $102 trillion in losses globally, equivalent to 1.05% of global GDP¡ªan alarming increase from $55.3 billion in 2022. Telecommunications, including SMS/MMS and voice/video calls, remain among the most widely used social applications, with a user base of around 9.3 billion, making them prime targets for online scams.

Fraudsters have employed various techniques to make their fraud attempts more complex and difficult to detect, including frequent changes in content, frequency, and ports. Telecom network fraud has become rampant, resulting in not only significant economic losses for the public but also negative social impacts. Traditional governance systems, based on static rules, have long business launch cycles, are unable to dynamically adjust to new fraud tactics, and have low recognition ability, efficiency, and increasing labor costs. These systems fall far short of meeting the growing need for effective governance.

Smart Safeguard is a pioneering AI MLLM-based anti-fraud system deployed in the telecommunications sector. It is capable of analyzing and processing various media types, including text, images, audio, and video, leveraging large language models, computer vision models, and hybrid multimodal models. The system provides robust multilingual support in languages such as English, French, Spanish, Chinese, Burmese, Hokkien, and Cantonese.

Proven to be intelligent, efficient, accurate, and comprehensive, the system features dynamic self-upgrading capabilities, offering a path forward for AI-driven governance in telecom network anti-fraud. It is easy to deploy, requiring no policy configuration, and fully automated workflows that perform data cleaning, import, training, and inference. It supports autonomous iterative closed-loop services, ensuring continuous updates to the system capabilities without manual intervention. The architecture uses a large model combined with load balancing, supporting high-capacity and high-concurrency service processing. Additionally, it adopts a mixed-size model with multi-layer serial and parallel deployment, improving performance while minimizing AI computing power costs.

The system ensures comprehensive security, robustness, and interoperability. Data security is maintained with the assurance that personal data never leaves the network, and privacy is protected. Security is also ensured in model training and deployment. For robustness, the system features a fully redundant architecture, and the business inference relies on load-balancing N+1 architecture for concurrent processing, ensuring extremely high reliability. With incremental fine-tuning capabilities, inference accuracy remains stable above 95%. In terms of interoperability, the system integrates seamlessly with any operator network through standard interfaces, serving users across the entire network.

Since its commercial launch in multiple provinces in China, the system has proven its ability to significantly enhance the efficiency, accuracy, and precision of content governance and anti-fraud efforts. It reduces the complexity of system management and the burden of manual review, providing substantial social and economic value.

In October 2023, China Unicom¡¯s Jiangsu Branch, in collaboration with ZTE, deployed the Smart Safeguard anti-fraud system. The system currently achieves a prediction accuracy and recall rate of over 95% for fraud detection, dramatically reducing the error rate compared to traditional systems. The results have been outstanding, reducing the economic losses of the public. The commercial deployment of the system has proven its ability to save costs in traditional system maintenance and manual review, while also reducing false positives, leading to increased revenue and comprehensive economic benefits.

Socially, this solution significantly enhances the user communication experience, providing worry-free services, protecting public property, and contributing to social stability. Additionally, the system supports an open API architecture, offering capability analysis APIs to both individual (ToC) and enterprise (ToB) users via an open gateway. It provides open capabilities and new business models for telecom networks and has been recommended for inclusion in the GSMA Open Gateway initiative.

The MLLM anti-fraud application of Smart Safeguard represents a technological breakthrough and a new application of AI large models in the communications industry. China Unicom and ZTE continue to lead in innovation and value. By using AI to govern AI, the telecommunications industry is evolving toward AI-powered, highly trusted networks, driving the evolution of communications toward 6G and beyond.