Alibaba's Qwen model easily modified for terrorist use
alibaba's qwen 3.8-27b model had guardrails removed within 24 hours of release, allowing terrorists to use it for harmful activities. this raises concerns about ai safety and regulation.
on august 14, chinese tech giant alibaba released qwen 3.8-27b, an open-weight artificial intelligence model. within 24 hours, versions of the model with removed guardrails, known as 'abliterated' models, appeared online. a neo-nazi, joehnny doomsayer, used an abliterated version to train a chatbot called 'mechahitler' and offered access to it for $14.88. researchers from tech against terrorism tested 134 ai models and found that most can be easily modified to assist terrorist and extremist activities. the researchers fed the models over 200 prompts covering topics such as mass casualty attacks and chemical weapons. only two models passed completely, while 81 failed, providing either complete answers or scores below 90.