Key Insights from the 2022 China AI Industry Conference on Security

The 2022 China AI Industry Conference (CAIIAM2022) was held on May 7‑8 at Jinji Lake, organized by the China Association for Artificial Intelligence. The event, themed “Scenario‑Driven, Intelligent Nation,” featured top scientists such as Academicians Dai Qionghai, Zheng Nanning, and Jiang Changjun, who shared the latest AI technologies and industry trends.

The AI Security Forum, chaired by researcher Lin Chenhao, gathered leading scholars and industry experts to discuss trustworthy, reliable, explainable, and fair AI, presenting key technologies, challenges, and application practices.

OPPO security expert Dr. Yang Minghui delivered a talk titled “Building Trustworthy Mobile Intelligent Terminals,” emphasizing the growing complexity of security in the era of ubiquitous connectivity and the need for comprehensive lifecycle protection covering privacy, data security, reliability, transparency, and accountability.

ColorOS, OPPO’s deeply customized Android system, implements a six‑layer security architecture—system, identity, connectivity, data, algorithm, and application—providing encryption, permission management, and privacy safeguards, and has earned multiple certifications such as CC MDFPP and Tier‑5 mobile terminal security assessments.

Professor Chen Kai from the Institute of Information Engineering, Chinese Academy of Sciences, presented “Adversarial and Defense of AI Models,” covering adversarial samples, model backdoors, fine‑grained repairs, defense methods, model stealing, and watermark protection.

Professor Wang Zhibo of Zhejiang University discussed “AI Security Risks and Trustworthy Exploration,” outlining risks in AI systems and recent work on adversarial samples, privacy attacks, and fairness enhancements.

Professor Ji Shouling, director of the Trusted AI Research Center, introduced the USENIX Security 2023 paper “FreeEagle: Detecting Complex Neutral Trojans in Data‑Free Cases,” a neural network trojan detector that works without data dependence and handles diverse complex backdoors.

Professor Liu Xianglong from Beihang University shared “Deep Learning Adversarial Attack and Defense Evaluation,” describing the team’s work on adversarial sample generation, backdoor detection, defense strategies, and the development of the “Zhongming” security evaluation platform and the “Wanxiang” physical simulation platform.

Ant Group senior algorithm expert Zhang Zhiqiang presented “Data Intelligence Practices in Technology‑Enabled Real‑World Applications,” illustrating how graph learning powers large‑scale deployments in digital life and digital finance.

A roundtable on “Large AI Models for Intelligent Security and Industry Challenges” highlighted ongoing opportunities and challenges related to data and privacy security, trustworthiness, ethics, and fairness, emphasizing the need for sustained research‑industry collaboration.