2024 WAIC Forum on Privacy Computing: Enabling Trusted Data Sharing for Large Models

High‑quality professional data flow and sharing are crucial for accelerating large‑model applications, yet the massive use of data introduces security and privacy challenges. Privacy computing—through encryption, desensitization, and anonymization—ensures data is not leaked during processing while enabling secure sharing and analysis, opening new opportunities for AI and big‑data development.

On July 5, the "2024 WAIC Privacy Computing: Assisting Trusted Fusion of Large Models and Data" forum was held under the guidance of the World AI Conference Committee and organized by Ant Group, China Telecom, China Academy of Information and Communications Technology, and Zhejiang University’s Blockchain and Data Security Laboratory.

Speakers, including renowned scholars and industry leaders, explored frontier privacy‑computing technologies, data‑trust standards, and the development of secure computation industries. Highlights included Dawn Song (UC Berkeley) emphasizing the importance of data for AI models and the role of privacy‑preserving computation, Ant Group’s VP Wei Tao discussing the dual nature of data value and associated risks, and Prof. Yu Yu (Shanghai Jiao Tong University) presenting a post‑quantum perspective on privacy computing.

Additional insights covered secure multi‑party computation’s quantum challenges, GLM’s data‑security practices (Dr. Gu Xiaotao, Zhipu AI), GDPR implications (Prof. Christoph Krönke), macro‑level data‑flow analysis (Chen Junyan, China Academy), and the need for unified standards (Yang Bo, China Academy).

Ant Group announced the launch of the "YinYu Cloud" large‑model privacy‑computing platform, offering encrypted model hosting and inference services to protect model assets, data security, and user privacy.

Round‑table discussions addressed large‑model data‑application challenges, evaluation metrics, and future directions, with participants stressing performance‑security balance, ecosystem building, and the vision of privacy computing becoming as seamless as network services.