Study warns of AI in terrorism
A study conducted by the "Technology Against Terrorism" initiative, supported by the United Nations, concluded that many AI models lack sufficient protective measures to prevent their use in requests related to carrying out terrorist operations. Out of 162 models tested, traditional models such as ChatGPT, Claude, and Gemini responded with only about 1.9% accuracy to questions indicating terrorist intent, while this rate could rise to 87-92% after modifications that disable security features. The findings suggest that open-source AI models can be easily altered and used without restrictions within a few days, increasing potential risks of their misuse for terrorist purposes.
articleDetails.fontSize.sizeحجم النص
تنويه: هذا ملخص تم إنشاؤه بواسطة الذكاء الاصطناعي
comments.heading