大模型易被恶意操控，安全治理迫在眉睫

Group 1 - The core issue highlighted is the frequent security vulnerabilities in large AI models, indicating that technological advancement must be accompanied by security measures [4] - The article discusses the risk of "data poisoning" and indirect prompt injection attacks, which can lead to the manipulation of model outputs and the potential theft of sensitive data [4] - It emphasizes that the security of large models is not merely a technical issue but a systemic challenge related to public safety, necessitating a proactive approach in model development, data training, and deployment [4] Group 2 - The industry is urged to prioritize security in the development and application of AI models to build robust defenses against potential threats [4] - The article suggests that a dual focus on technology and security is essential to prevent AI from "losing control" during rapid advancements [4]