The Security Nakedness Behind the Frenzy of Top Chinese Open-Source Large Models: 76% Vulnerability Reproduction Rate, Zero Refusal of High-Risk Instructions - 36 Kr
A recent study reveals that China's top open-source large models have severe security flaws, with a 76% vulnerability reproduction rate and no refusal to high-risk instructions. This raises concerns about the safety of these models in real-world applications.
A report from 36 Kr highlights critical security issues in China's leading open-source large models. The study found that 76% of known vulnerabilities could be reproduced in these models, and they showed zero refusal to high-risk instructions. This means that the models are highly susceptible to malicious use, posing significant risks if deployed in real-world scenarios. The findings underscore the urgent need for enhanced security measures in the development and deployment of open-source AI models.
Source: Zhipu GLM (GNews) —
original