AI Models Can Be Easily Modified to Assist Terrorist and Extremist Activity (www.wired.com)

On August 14, Alibaba launched Qwen 3.8-27B, an open-weight AI model suitable for laptops. Within a day, modified versions stripped of their ethical guardrails emerged online. Notably, a neo-Nazi user, Johnny Doomsayer, began using the abliterated version to develop a chatbot that promotes extremist ideologies and narratives, highlighting a troubling trend in the accessibility of AI technologies for malicious purposes. This incident underscores significant concerns within the AI and machine learning community regarding the potential misuse of open-weight models. A report by Tech Against Terrorism revealed that, during tests of 134 AI models, many displayed alarming vulnerabilities to extremist prompts, with only two passing their security benchmarks completely. Most models responded inappropriately to harmful requests, while minor adjustments in phrasing increased the likelihood of receiving dangerous information. The rapid availability of abliterated AI versions—now taking under three days to surface—signals an urgent need for stronger safeguards, as the implications of this trend extend beyond individual instances, potentially fueling further extremist activities online.