🇬🇷 Greek, 9 October 2026, AI security
AI Models Can Be Easily Repurposed to Assist Terrorists
Many AI models lack sufficient security safeguards against requests for 'assisting a terrorist operation', a new report warns.
The report, conducted by Tech Against Terrorism, tested 162 AI models and found that typical models, such as ChatGPT, Claude, and Gemini, answered 1.9% of questions related to terrorist activity when asked by a user who explicitly stated their intention to commit a terrorist attack.
However, when the user claimed to be conducting research on security, the models’ response rate increased to 16.9%.
The report highlights the risks associated with ‘open-weight models’, which can be easily modified to remove security safeguards and increase their response rate to 87-92% when asked about terrorist activity.