According to WIRED, a new report from Tech Against Terrorism finds that open-weight AI models can have their guardrails stripped and be shared online within days of release. The report cites Alibaba's Qwen 3.8-27B, which was released on August 14 and appeared in "abliterated" form within 24 hours. Three days later, a neo-Nazi known as Johnny Doomsayer said he was using that version to train a chatbot called MechaHitler, advertising access for $14.88.
The report tested 134 AI models with more than 200 prompts covering topics such as mass-casualty attacks and weapons production. Most closed-weight models from major US companies performed well overall, but 81 models failed, meaning they either gave a complete answer on a mass-casualty subject or scored below 90. Only two models passed completely. The researchers also found that reframing a request as coming from a "safety researcher" made models almost nine times as likely to provide an answer, with a 16.9 percent compliance rate.
The researchers describe an "abliteration emergency" that is getting worse. On average, abliterated versions of newly released open-weight models now appear in less than three days, compared with more than 24 days in 2024. Meta's Llama 3.1 8B scored 97 out of 100 on the benchmark, while an abliterated version scored 3. When asked how to build an improvised explosive device, the abliterated model responded with 15 numbered steps under three headings, ending with a section titled "Important Safety Considerations." Meta, OpenAI, Google, Anthropic, xAI, and Alibaba did not respond to requests for comment on the findings.